JP2004061558A - Method and device for code conversion between speed encoding and decoding systems and storage medium therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device and a method that convert codes obtained by encoding a speech by a certain system into codes that another system can decode with high tone quality and a small computation quantity. <P>SOLUTION: In a code converting device which converts a 1st code sequence based upon a 1st system into a 2nd code sequence based upon a 2nd system, a speech decoding circuit (1500) obtains information on a 1st linear prediction coefficient and an exciting signal from the 1st code sequence and drives a filter having the 1st linear prediction coefficient with an exciting signal obtained from the information of the exciting signal to generate a 1st speech signal, and a gain code generating circuit (1400) computes such a gain (optimum gain) that the distance between a 1st speech signal generated with the information obtained from the 2nd code sequence and the 1st speech signal becomes minimum, corrects the optimum gain, and finds gain information on the 2nd code sequence according to the corrected optimum gain, the optimum gain, and a gain read out of a gain code book of the 2nd system. <P>COPYRIGHT: (C)2004,JPO

Description

ｍｓｅｃ周期（フレーム）毎に行われ、ＡＣＢ、ＦＣＢ及びゲインなど励振信号の構成要素の符号化は、

ｍｓｅｃ周期（サブフレーム）毎に行われるものとする。
【００５６】
一方、方式Ｂにおいては、ＬＰ係数の符号化は、

ｍｓｅｃ周期（フレーム）毎に行われ、励振信号の構成要素の符号化は、

ｍｓｅｃ周期（サブフレーム）毎に行われるものとする。
【００５７】
また、方式Ａのフレーム長、サブフレーム数、及びサブフレーム長を、それぞれ、

、

及び

とする。
【００５８】
方式Ｂのフレーム長、サブフレーム数、及び、サブフレーム長を、それぞれ、

、

及び、

とする。
【００５９】
以下の説明では、簡単のため、

とする。
【００６０】
ここで、例えば、サンプリング周波数を、８０００Ｈｚとし、

及び

を１０　ｍｓｅｃとすれば、

及び

は１６０サンプル、

及び

は８０サンプルとなる。
【００６１】
ＬＰ係数符号変換回路１１００は、符号分離回路１０１０から第１のＬＰ係数符号を入力する。ここで、「３ＧＰＰ　ＡＭＲ　Ｓｐｅｅｃｈ　Ｃｏｄｅｃ」（文献３）や、ＩＴＵ−Ｔ勧告Ｇ．７２９など多くの標準方式では、ＬＰ係数を線スペクトル対（Ｌｉｎｅ　Ｓｐｅｃｔｒａｌ　Ｐａｉｒ：　ＬＳＰ）で表現し、ＬＳＰを符号化及び復号することが多いため、ＬＰ係数の符号化及び復号は、ＬＳＰ領域で行われるとする。ＬＰ係数からＬＳＰへの変換、及びＬＳＰからＬＰ係数への変換については、周知の方法、例えば「文献３」の第５．２．３節及び第５．２．４節の記載が参照される。ＬＰ係数符号変換回路１１００は、前記第１のＬＰ係数符号を方式ＡにおけるＬＳＰ復号方法により復号して、第１のＬＳＰを得る。
【００６２】
次に、ＬＰ係数符号変換回路１１００は、前記第１のＬＳＰを、方式ＢにおけるＬＳＰ量子化方法及び符号化方法により量子化及び符号化して、第２のＬＳＰとこれに対応する符号（第２のＬＰ係数符号）を得る。そして、ＬＰ係数符号変換回路１１００は、前記第２のＬＰ係数符号を方式ＢにおけるＬＳＰ復号方法により復号可能な符号として符号多重回路１０２０へ出力し、前記第１のＬＳＰと第２のＬＳＰをＬＳＰ−ＬＰＣ変換回路１１１０へ出力する。
【００６３】
図２は、ＬＰ係数符号変換回路１１００の構成を示す図である。図２を参照すると、ＬＰ係数符号変換回路１１００は、ＬＳＰ復号回路１１０と、第１のＬＳＰコードブック１１１と、ＬＳＰ係数符号化回路１３０と、第２のＬＳＰコードブック１３１とを備えている。図２を参照して、ＬＰ係数符号変換回路１１００の各構成要素について説明する。
【００６４】
ＬＳＰ復号回路１１０は、ＬＰ係数符号から対応するＬＳＰを復号する。ＬＳＰ復号回路１１０は、複数セットのＬＳＰが格納された第１のＬＳＰコードブック１１１を備えており、符号分離回路１０１０から出力される第１のＬＰ係数符号を、入力端子３１を介して入力し、第１のＬＰ係数符号に対応するＬＳＰを第１のＬＳＰコードブック１１１より読み出し、読み出されたＬＳＰを第１のＬＳＰとしてＬＳＰ符号化回路１３０へ出力するとともに、出力端子３３を介してＬＳＰ−ＬＰＣ変換回路１１１０へ出力する。ここで、ＬＰ係数符号からのＬＳＰの復号は、方式ＡにおけるＬＳＰの復号方法に従い、方式ＡのＬＳＰコードブックを用いる。
【００６５】
ＬＳＰ符号化回路１３０は、ＬＳＰ復号回路１１０から出力される第１のＬＳＰを入力し、複数セットのＬＳＰが格納された第２のＬＳＰコードブック１３１から第２のＬＳＰとそれに対応するＬＰ係数符号の各々を順次読み込み、第１のＬＳＰとの誤差が最小となる第２のＬＳＰを選択し、それに対応するＬＰ係数符号を、第２のＬＰ係数符号として出力端子３２を介して符号多重回路１０２０へ出力し、第２のＬＳＰを出力端子３４を介してＬＳＰ−ＬＰＣ変換回路１１１０へ出力する。ここで、第２のＬＳＰの選択方法、すなわちＬＳＰの量子化及び符号化方法は、方式ＢにおけるＬＳＰの量子化方法及び符号化方法に従い、方式ＢのＬＳＰコードブックを用いる。ここで、ＬＳＰの量子化及び符号化については、例えば「文献３」の第５．２．５節の記載が参照される。
【００６６】
以上により、図２によるＬＰ係数符号変換回路１１００の説明を終え、再び図１の説明に戻る。
【００６７】
ＬＳＰ−ＬＰＣ変換回路１１１０は、ＬＰ係数符号変換回路１１００から出力される第１のＬＳＰと第２のＬＳＰとを入力し、第１のＬＳＰを第１のＬＰ係数ａ_１，ｉに変換し、第２のＬＳＰを第２のＬＰ係数ａ_２，ｉに変換し、第１のＬＰ係数ａ_１，ｉを目標信号計算回路１７００と、音声復号回路１５００と、インパルス応答計算回路１１２０へ出力し、第２のＬＰ係数ａ_２，ｉを目標信号計算回路１７００とインパルス応答計算回路１１２０へ出力する。ここで、ＬＳＰからＬＰ係数への変換については、「文献３」の第５．２．４節の記載が参照される。
【００６８】
ＡＣＢ符号変換回路１２００は、符号分離回路１０１０から入力した第１のＡＣＢ符号を、方式Ａにおける符号と方式Ｂにおける符号との対応関係を用いて読み替えることにより、第２のＡＣＢ符号を得る。そして、ＡＣＢ符号変換回路１２００は、第２のＡＣＢ符号を方式ＢにおけるＡＣＢ復号方法により復号可能な符号として符号多重回路１０２０へ出力する。また、ＡＣＢ符号変換回路１２００は、第２のＡＣＢ符号に対応するＡＣＢ遅延を第２のＡＣＢ遅延として目標信号計算回路１７００へ出力する。
【００６９】
ここで、図３を参照して、符号の読み替えについて説明する。例えば、方式ＡにおけるＡＣＢ符号

が５６のとき、これに対応するＡＣＢ遅延

が７６であるとする。方式Ｂでは、ＡＣＢ符号

が５３のとき、これに対応するＡＣＢ遅延

が７６であるとすると、ＡＣＢ遅延の値が同一（この場合では７６）となるように、方式Ａから方式ＢへとＡＣＢ符号を変換するには、方式ＡにおけるＡＣＢ符号５６を方式ＢにおけるＡＣＢ符号５３に対応付ければよい。以上により、符号の読み替えについての説明を終え、再び図１の説明に戻る。
【００７０】
音声復号回路１５００は、符号分離回路１０１０から出力される第１のＡＣＢ符号、第１のＦＣＢ符号、第１のゲイン符号を入力し、ＬＳＰ−ＬＰＣ変換回路１１１０から第１のＬＰ係数を入力する。次に、音声復号回路１５００は、方式Ａにおける、ＡＣＢ信号復号方法、ＦＣＢ信号復号方法及びゲイン復号方法の各々を用いて、第１のＡＣＢ符号、第１のＦＣＢ符号及び第１のゲイン符号の各々から、ＡＣＢ遅延、ＦＣＢ信号及びゲインの各々を復号し、各々を第１のＡＣＢ遅延、第１のＦＣＢ信号及び第１のゲインとする。音声復号回路１５００は、第１のＡＣＢ遅延を用いてＡＣＢ信号を生成し、これを第１のＡＣＢ信号とする。そして、音声復号回路１５００は、第１のＡＣＢ信号、第１のＦＣＢ信号及び第１のゲインと、第１のＬＰ係数とから、音声を生成し、音声を目標信号計算回路１７００へ出力する。
【００７１】
図４は、音声復号回路１５００の構成を示す図である。図４を参照すると、音声復号回路１５００は、ＡＣＢ復号回路１５１０と、ＦＣＢ復号回路１５２０と、ゲイン復号回路１５３０とを有する励振信号情報復号回路１６００と、励振信号計算回路１５４０と、励振信号記憶回路１５７０と、合成フィルタ１５８０を備えている。図４を参照して、音声復号回路１５００の各構成要素について説明する。
【００７２】
励振信号情報復号回路１６００は、励振信号の情報に対応する符号から励振信号の情報を復号する。符号分離回路１０１０から出力される第１のＡＣＢ符号、第１のＦＣＢ符号及び第１のゲイン符号を各々入力端子５１、５２及び５３を介して入力し、第１のＡＣＢ符号、第１のＦＣＢ符号及び第１のゲイン符号の各々から、ＡＣＢ遅延、ＦＣＢ信号及びゲインの各々を復号し、各々を第１のＡＣＢ遅延、第１のＦＣＢ信号及び第１のゲインとする。ここで、第１のゲインは、ＡＣＢゲインとＦＣＢゲインとからなり、各々を第１のＡＣＢゲインと第１のＦＣＢゲインとする。また、励振信号情報復号回路１６００は、励振信号記憶回路１５７０から出力される過去の励振信号を入力する。励振信号情報復号回路１６００は、過去の励振信号と第１のＡＣＢ遅延とを用いてＡＣＢ信号を生成し、これを第１のＡＣＢ信号とする。そして、励振信号情報復号回路１６００は、第１のＡＣＢ信号、第１のＦＣＢ信号、第１のＡＣＢゲイン及び第１のＦＣＢゲインを、励振信号計算回路１５４０へ出力する。
【００７３】
次に、励振信号情報復号回路１６００の構成要素であるＡＣＢ復号回路１５１０、ＦＣＢ復号回路１５２０、及びゲイン復号回路１５３０について詳細に説明する。
【００７４】
ＡＣＢ復号回路１５１０は、符号分離回路１０１０から出力される第１のＡＣＢ符号を、入力端子５１を介して入力し、励振信号記憶回路１５７０から出力される過去の励振信号を入力する。次に、ＡＣＢ復号回路１５１０は、上述したＡＣＢ符号変換回路１２００と同様にして、図３に示す方式ＡにおけるＡＣＢ　符号とＡＣＢ遅延の対応関係を用いて、第１のＡＣＢ　符号に対応する第１のＡＣＢ遅延

を得る。励振信号において、現サブフレームの始点より

サンプル過去の点から、サブフレーム長に相当する

サンプルの信号を切り出して、第１のＡＣＢ信号を生成する。ここで、

が

よりも小さい場合には、

サンプル分のベクトルを切り出し、このベクトルを繰り返し接続して、長さ

サンプルの信号とする。そして、第１のＡＣＢ信号を励振信号計算回路１５４０へ出力する。ここで、第１のＡＣＢ信号を生成する方法の詳細については、「文献３」の第６．１節及び第５．６節の記載が参照される。
【００７５】
ＦＣＢ復号回路１５２０は、符号分離回路１０１０から出力される第１のＦＣＢ符号を、入力端子５２を介して入力し、第１のＦＣＢ符号に対応する第１のＦＣＢ信号を、励振信号計算回路１５４０へ出力する。ＦＣＢ信号は、パルス位置とパルス極性で規定されるマルチパルス信号により表現されており、第１のＦＣＢ符号はパルス位置に対応する符号（パルス位置符号）とパルス極性に対応する符号（パルス極性符号）とからなる。ここで、マルチパルス信号により表現されたＦＣＢ信号を生成する方法の詳細については、「文献３」の第６．１節及び第５．７節の記載が参照される。
【００７６】
ゲイン復号回路１５３０は、符号分離回路１０１０から出力される第１のゲイン符号を、入力端子５３を介して入力する。ゲイン復号回路１５３０は、複数のゲインが格納されたテーブルを内蔵しており、第１のゲイン符号に対応するゲインをテーブルから読み出す。そして、ゲイン復号回路１５３０は、読み出されたゲインのうち、ＡＣＢゲインに対応する第１のＡＣＢゲインと、ＦＣＢゲインに対応する第１のＦＣＢゲインとを励振信号計算回路１５４０へ出力する。ここで、第１のＡＣＢゲインと第１のＦＣＢゲインがまとめて符号化されている場合には、テーブルには第１のＡＣＢゲインと第１のＦＣＢゲインとから成る２次元ベクトルが複数格納されている。また、第１のＡＣＢゲインと第１のＦＣＢゲインが個別に符号化されている場合には、二つのテーブルが内蔵され、一方のテーブルに第１のＡＣＢゲインが複数格納されており、他方のテーブルに第１のＦＣＢゲインが複数格納されている。
【００７７】
励振信号計算回路１５４０は、ＡＣＢ復号回路１５１０から出力される第１のＡＣＢ信号を入力し、ＦＣＢ復号回路１５２０から出力される第１のＦＣＢ信号を入力し、ゲイン復号回路１５３０から出力される第１のＡＣＢゲインと第１のＦＣＢゲインとを入力する。励振信号計算回路１５４０は、第１のＡＣＢ信号に第１のＡＣＢゲインを乗じて得た信号と、第１のＦＣＢ信号に第１のＦＣＢゲインを乗じて得た信号とを加算して第１の励振信号を得る。そして、励振信号計算回路１５４０は、第１の励振信号を、合成フィルタ１５８０と励振信号記憶回路１５７０とへ出力する。
【００７８】
励振信号記憶回路１５７０は、励振信号計算回路１５４０から出力される第１の励振信号を入力し、これを記憶保持する。そして、励振信号記憶回路１５７０は、過去に入力されて記憶保持されている過去の第１の励振信号をＡＣＢ復号回路１５１０へ出力する。
【００７９】
合成フィルタ１５８０は、励振信号計算回路１５４０から出力される第１の励振信号を入力し、ＬＳＰ−ＬＰＣ変換回路１１１０から出力される第１のＬＰ係数を入力端子６１を介して入力する。そして、合成フィルタ１５８０は、第１のＬＰ係数をもつ線形予測フィルタを、第１の励振信号で駆動することにより音声信号を生成する。音声信号を目標信号計算回路１７００へ出力端子６３を介して出力する。
【００８０】
以上で、図４による音声復号回路１５００の説明を終え、再び図１の説明に戻る。
【００８１】
目標信号計算回路１７００は、ＬＳＰ−ＬＰＣ変換回路１１１０から第１のＬＳＰと第２のＬＳＰとを入力し、ＡＣＢ符号変換回路１２００から第２のＡＣＢ符号に対応する第２のＡＣＢ遅延を入力し、音声復号回路１５００から復号音声を入力し、インパルス応答計算回路１１２０からインパルス応答信号を入力し、第２の励振信号記憶回路１６２０に記憶保持される過去の第２の励振信号を入力する。目標信号計算回路１７００は、復号音声と第１のＬＰ係数及び第２のＬＰ係数とから第１の目標信号を計算する。次に、目標信号計算回路１７００は、過去の第２の励振信号とインパルス応答信号と第１の目標信号と第２のＡＣＢ遅延とから、第２のＡＣＢ信号及び最適ＡＣＢゲインを求める。そして、目標信号計算回路１７００は、第１の目標信号と最適ＡＣＢゲインとをゲイン符号生成回路１４００へ出力し、第２のＡＣＢ信号をゲイン符号生成回路１４００と第２の励振信号計算回路１６１０とへ出力する。
【００８２】
図５は、目標信号計算回路１７００の構成を示す図である。図５を参照すると、目標信号計算回路１７００は、重み付け信号計算回路１７１０と、ＡＣＢ信号生成回路１７２０と、最適ＡＣＢゲイン計算回路１７３０とを備えている。図５を参照して、目標信号計算回路１７００の各構成要素について説明する。
【００８３】
重み付け信号計算回路１７１０は、音声復号回路１５００の合成フィルタ１５８０から出力される復号音声ｓ（ｎ）を入力端子５７を介して入力し、ＬＳＰ−ＬＰＣ変換回路１１１０から出力される第１のＬＰ係数ａ_１，ｉと第２のＬＰ係数ａ_２，ｉとを、各々入力端子３６と入力端子３５とを介して入力する。重み付け信号計算回路１７１０は、まず、第１のＬＰ係数を用いて、聴感重み付けフィルタＷ（ｚ）を構成する。
【００８４】
そして、重み付け信号計算回路１７１０は、復号音声により聴感重み付けフィルタを駆動して聴感重み付け音声信号を生成する。次に、重み付け信号計算回路１７１０は、第１のＬＰ係数と第２のＬＰ係数とを用いて、聴感重み付け合成フィルタＷ（ｚ）／Ａ２（ｚ）を構成する。
【００８５】
そして、重み付け信号計算回路１７１０は、聴感重み付け合成フィルタの零入力応答を聴感重み付け音声信号から減算して得られる第１の目標信号ｘ（ｎ）を、ＡＣＢ信号生成回路１７２０と最適ＡＣＢゲイン計算回路１７３０へ出力するとともに、第２の目標信号計算回路１４３０へ出力端子７８を介して出力する。
【００８６】
ＡＣＢ信号生成回路１７２０は、重み付け信号計算回路１７１０から出力される第１の目標信号を入力し、ＡＣＢ符号変換回路１２００から出力される第２のＡＣＢ遅延Ｔ^（Ｂ） _ｌａｇを入力端子３７を介して入力し、インパルス応答計算回路１１２０から出力されるインパルス応答信号ｈ（ｎ）を入力端子７４を介して入力し、第２の励振信号記憶回路１６２０から出力される過去の第２の励振信号ｕ（ｎ）を入力端子７５を介して入力する。
【００８７】
ＡＣＢ信号生成回路１７２０は、過去の第２の励振信号から遅延ｋで切り出された信号とインパルス応答信号との畳み込みにより、フィルタ処理された遅延ｋの過去の励振信号

を計算する。
【００８８】
ここで、遅延ｋは第２のＡＣＢ遅延とする。過去の第２の励振信号から遅延ｋで切り出された信号を第２のＡＣＢ信号ｖ（ｎ）とする。
【００８９】
そして、ＡＣＢ信号生成回路１７２０は、第２のＡＣＢ信号を第２の目標信号計算回路１４３０と第２の励振信号計算回路１６１０とへ出力端子７６を介して出力し、フィルタ処理された遅延ｋの過去の励振信号ｙｋ（ｎ）を最適ＡＣＢゲイン計算回路１７３０へ出力する。
【００９０】
最適ＡＣＢゲイン計算回路１７３０は、重み付け信号計算回路１７１０から出力される第１の目標信号ｘ（ｎ）を入力し、ＡＣＢ信号生成回路１７２０から出力されるフィルタ処理された遅延ｋの過去の励振信号ｙｋ（ｎ）を入力する。
【００９１】
次に、最適ＡＣＢゲイン計算回路１７３０は、第１の目標信号ｘ（ｎ）と、フィルタ処理された遅延ｋの過去の励振信号ｙｋ（ｎ）と、から最適ＡＣＢゲインｇｐを次式により計算する。最適ＡＣＢゲインｇｐは、第１の目標信号ｘ（ｎ）と、フィルタ処理された遅延ｋの過去の励振信号ｙｋ（ｎ）との距離を最小とするゲインである。

【００９２】
そして、最適ＡＣＢゲイン計算回路１７３０は、最適ＡＣＢゲインｇｐをＡＣＢゲイン符号化回路１４１０へ出力端子７７を介して出力する。
【００９３】
なお、第２のＡＣＢ信号を計算する方法及び最適ＡＣＢゲインを計算する方法の詳細については、「文献３」の第６．１節及び第５．６節の記載が参照できる。以上で図５による目標信号計算回路１７００の説明を終え、再び図１の説明に戻る。
【００９４】
インパルス応答計算回路１１２０は、ＬＳＰ−ＬＰＣ変換回路１１１０から出力される第１のＬＰ係数と第２のＬＰ係数を入力し、第１のＬＰ係数と第２のＬＰ係数を用いて聴感重み付け合成フィルタを構成する。
【００９５】
そして、インパルス応答計算回路１１２０は、聴感重み付け合成フィルタのインパルス応答信号を目標信号計算回路１７００とゲイン符号生成回路１４００とへ出力する。ここで、聴感重み付け合成フィルタの伝達関数は次式により表される。

【００９６】
ただし、

【００９７】
は、第２のＬＰ係数

をもつ線形予測フィルタの伝達関数である。
【００９８】

【００９９】
は、第１のＬＰ係数

をもつ聴感重み付けフィルタの伝達関数である。
【０１００】
ここで、Ｐは、線形予測次数（例えば、１０）であり、γ１とγ２は、重み付けを制御する係数（例えば、０．９４と０．６）である。
【０１０１】
ＦＣＢ符号生成回路１８００は、符号分離回路１０１０から出力される第１のＦＣＢ符号を入力し、第１のＦＣＢ符号を方式Ｂにより復号可能な符号に変換する。ＦＣＢ符号生成回路１８００は、変換されたＦＣＢ符号を、第２のＦＣＢ符号として符号多重回路１０２０へ出力し、第２のＦＣＢ符号に対応する第２のＦＣＢ信号をゲイン符号生成回路１４００と、第２の励振信号計算回路１６１０とへ出力する。ここで、ＦＣＢ信号は、複数のパルスから成り、パルスの位置（パルス位置）と極性（パルス極性）で規定されるマルチパルス信号により表現される。ＦＣＢ符号は、パルス位置に対応する符号（パルス位置符号）とパルス極性に対応する符号（パルス極性符号）とからなる。マルチパルス信号によるＦＣＢ信号の表現方法については、「文献３」の第５．７節の記載が参照される。
【０１０２】
図６は、図１のＦＣＢ符号生成回路１８００の構成を示す図である。図６を参照すると、ＦＣＢ符号生成回路１８００は、ＦＣＢ符号変換回路１３００と、ＦＣＢ信号生成回路１８２０を備えている。図６を参照して、ＦＣＢ符号生成回路１８００の各構成要素について説明する。
【０１０３】
ＦＣＢ符号変換回路１３００は、符号分離回路１０１０から入力端子８５を介して入力した第１のＦＣＢ符号ｉ^（Ａ） _Ｐを、方式Ａにおける符号と方式Ｂにおける符号との対応関係を用いて読み替えることにより、第２のＦＣＢ符号ｉ^{（ Ｂ ）} _Ｐを得る。そして、ＦＣＢ符号変換回路１３００は、これを方式ＢにおけるＦＣＢ復号方法により復号可能な符号として出力端子５５を介して符号多重回路１０２０へ出力し、第２のＦＣＢ符号に対応するパルス位置

及び、パルス極性

をＦＣＢ信号生成回路１８２０へ出力する。
【０１０４】
図７を参照して、パルス位置符号の読み替えについて説明する。
【０１０５】
例えば、方式Ａにおけるパルス位置符号

が６のとき、これに対応するパルス位置

が３０であるとする。方式Ｂでは、パルス位置符号

が１のとき、これに対応するパルス位置

が３０であるとすると、パルス位置の値が同一（この場合では３０）となるように、方式Ａから方式Ｂへとパルス位置符号を変換するには、方式Ａにおけるパルス位置符号６を方式Ｂにおけるパルス位置符号１に対応付ければよい。
【０１０６】
パルス極性符号については、読み替え前の符号に対応する極性（正又は負）と、読み替え後の符号に対応する極性とが等しくなるように、符号を読み替えればよい。
【０１０７】
以上により、パルス位置符号及びパルス極性符号の読み替えについての説明を終え、再び図６の説明に戻る。
【０１０８】
ＦＣＢ信号生成回路１８２０は、ＦＣＢ符号変換回路１３００から出力されるパルス位置及びパルス極性を入力する。ＦＣＢ信号生成回路１８２０は、パルス位置及びパルス極性から規定されるＦＣＢ信号を第２のＦＣＢ信号ｃ（ｎ）とし、これを最適ＦＣＢゲイン計算回路１４４０と第２の励振信号計算回路１６１０とへ出力端子８６を介して出力する。
【０１０９】
以上で図６によるＦＣＢ符号生成回路１８００の説明を終え、再び図１の説明に戻る。
【０１１０】
ゲイン符号生成回路１４００は、目標信号計算回路１７００から出力される第１の目標信号と第２のＡＣＢ信号と最適ＡＣＢゲインとを入力し、ＦＣＢ符号生成回路１８００から出力される第２のＦＣＢ信号を入力し、インパルス応答計算回路１１２０から出力されるインパルス応答信号を入力し、ＬＰ係数符号変換回路１１００から出力される第１のＬＳＰを入力する。
【０１１１】
ゲイン符号生成回路１４００は、まず、第１の目標信号と第２のＡＣＢ信号と最適ＡＣＢゲインとインパルス応答信号とから第２の目標信号を計算し、第２の目標信号と第２のＦＣＢ信号とインパルス応答信号とから最適ＦＣＢゲインを計算し、最適ＦＣＢゲインから修正ＦＣＢゲインを計算し、第１のＬＳＰから音声判定値を決定する。
【０１１２】
次に、ゲイン符号生成回路１４００は、ＡＣＢゲインコードブックから順次読み込まれるＡＣＢゲインと最適ＡＣＢゲインとから第１の自乗誤差を計算し、ＡＣＢゲインと修正ＡＣＢゲインとから第２の自乗誤差を計算する。
【０１１３】
そして、ゲイン符号生成回路１４００は、音声判定値から計算される重み係数と第１の自乗誤差と第２の自乗誤差とから計算される評価関数が最小となるＡＣＢゲイン及び対応するＡＣＢゲイン符号を選択する。
【０１１４】
また、ゲイン符号生成回路１４００は、ＦＣＢゲインコードブックから順次読み込まれるＦＣＢゲインと最適ＦＣＢゲインとから第３の自乗誤差を計算し、ＦＣＢゲインと修正ＦＣＢゲインとから第４の自乗誤差を計算する。
【０１１５】
そして、ゲイン符号生成回路１４００は、音声判定値から計算される重み係数と第３の自乗誤差と第４の自乗誤差とから計算される評価関数が最小となるＦＣＢゲイン及び対応するＦＣＢゲイン符号を選択する。
【０１１６】
最後に、ゲイン符号生成回路１４００は、選択されたＡＣＢゲイン符号とＦＣＢゲイン符号とからなる第２のゲイン符号を、方式Ｂにおけるゲイン復号方法により復号可能な符号として符号多重回路１０２０へ出力端子５６を介して出力する。
【０１１７】
図８は、ゲイン符号生成回路１４００の構成を示す図である。図８を参照すると、ゲイン符号生成回路１４００は、ＡＣＢゲイン符号化回路１４１０と、ＡＣＢゲインコードブック１４１１と、ＦＣＢゲイン符号化回路１４２０と、ＦＣＢゲインコードブック１４２１と、第２の目標信号計算回路１４３０と、最適ＦＣＢゲイン計算回路１４４０と、最適ＦＣＢゲイン修正回路１４５０と、音声／非音声識別回路１４６０と、を備えている。図８を参照して、ゲイン符号生成回路１４００の各構成要素について詳細に説明する。
【０１１８】
第２の目標信号計算回路１４３０は、ＡＣＢ信号生成回路１７２０から出力される第２のＡＣＢ信号ｖ（ｎ）を入力端子９２を介して入力し、重み付け信号計算回路１７１０から出力される第１の目標信号ｘ（ｎ）を入力端子９３を介して入力し、インパルス応答計算回路１１２０から出力されるインパルス応答信号ｈ（ｎ）を入力端子９４を介して入力し、ＡＣＢゲイン符号化回路１４１０から出力される第２のＡＣＢゲインを入力する。
【０１１９】
第２の目標信号計算回路１４３０は、第２のＡＣＢ信号とインパルス応答信号との畳み込みにより、フィルタ処理された第２のＡＣＢ信号

を計算し、ｙ（ｎ）に第２のＡＣＢゲイン

を乗じて得られる信号を、第１の目標信号ｘ（ｎ）から減算して、第２の目標信号ｘ_２（ｎ）を得る。

【０１２０】
そして、第２の目標信号計算回路１４３０は、第２の目標信号ｘ_２（ｎ）を最適ＦＣＢゲイン計算回路１４４０へ出力する。
【０１２１】
最適ＦＣＢゲイン計算回路１４４０は、ＦＣＢ信号生成回路１８２０から出力される第２のＦＣＢ信号ｃ（ｎ）を入力端子９１を介して入力し、インパルス応答計算回路１１２０から出力されるインパルス応答信号ｈ（ｎ）を入力端子９４を介して入力し、第２の目標信号計算回路１４３０から出力される第２の目標信号ｘ_２（ｎ）を入力し、第２のＦＣＢ信号とインパルス応答信号との畳み込みによりフィルタ処理された第２のＦＣＢ信号

を計算し、第２の目標信号ｘ２（ｎ）とフィルタ処理された第２のＦＣＢ信号ｚ（ｎ）から、次の式により最適ＦＣＢゲインｇｃを計算する。最適ＦＣＢゲインｇｃは、第２の目標信号ｘ２（ｎ）とフィルタ処理された第２のＦＣＢ信号ｚ（ｎ）との距離を最小とするゲインである。

【０１２２】
そして、最適ＦＣＢゲイン計算回路１４４０は、最適ＦＣＢゲインを最適ＦＣＢゲイン修正回路１４５０とＦＣＢゲイン符号化回路１４２０とへ出力する。
【０１２３】
音声／非音声識別回路１４６０は、ＬＳＰ復号回路１１０から出力される第１のＬＳＰを入力端子９８を介して入力する。第１のＬＳＰとその長時間平均とからＬＳＰ変動量を計算し、ＬＳＰ変動量から音声判定値を決定する。
【０１２４】
ＬＳＰ変動量を求める手順を以下に示す。第ｎフレームにおいて、ＬＳＰの長時間平均

を次式により計算する。

ここで、Ｎｐは線形予測次数であり、βは例えば０．９である。
【０１２５】
第ｎフレームにおけるＬＳＰの変動量ｄｑ（ｎ）を次式により定義する。

ここで、

は、

と

との誤差として、例えば、

又は、

などが定義できるが、ここでは、後者を用いる。変動量ｄｑ（ｎ）が大きい区間を音声区間に、小さい区間を非音声区間に対応させることができる。変動量ｄｑ（ｎ）に対する閾値処理により、音声判定値

を決定する。
【０１２６】

（Ｖｓ＝１　ｄｑ（ｎ）がＣＶＳ以上の場合
Ｖｓ＝０　ｄｑ（ｎ）がＣＶＳより小の場合）
【０１２７】
ここで、Ｃｖｓはある定数（例えば、２．２）であり、Ｖｓ＝１は音声区間に、Ｖｓ＝０は非音声区間に対応する。音声判定値を最適ＡＣＢゲイン修正回路１４８０とＡＣＢゲイン符号化回路１４１０と最適ＦＣＢゲイン修正回路１４５０とＦＣＢゲイン符号化回路１４２０とへ出力する。
【０１２８】
最適ＡＣＢゲイン修正回路１４８０は、ＡＣＢ信号生成回路１７２０から出力される最適ＡＣＢゲインを入力端子９７を介して入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。最適ＡＣＢゲイン修正回路１４８０では、音声判定値Ｖｓが０（非音声区間）のとき、最適ＡＣＢゲインの長時間平均を修正ＡＣＢゲインとする。非音声区間において、次式により最適ＡＣＢゲインの長時間平均を計算する。

【０１２９】
ここで、

は第ｎサブフレームにおける最適ＡＣＢゲイン、

は第ｎサブフレームにおける最適ＡＣＢゲインの長時間平均であり、αは例えば０．９である。なお、長時間平均には平均値、中央値、最頻値なども適用できる。
【０１３０】
一方、最適ＡＣＢゲイン修正回路１４８０では、音声判定値Ｖｓが１（音声区間）のとき、最適ＡＣＢゲインそのものを修正ＡＣＢゲインとする。
【０１３１】
最適ＡＣＢゲイン修正回路１４８０は、修正ＡＣＢゲインを、ＡＣＢゲイン符号化回路１４１０へ出力する。
【０１３２】
ＡＣＢゲイン符号化回路１４１０は、ＡＣＢ信号生成回路１７２０から出力される最適ＡＣＢゲインｇｐを入力端子９７を介して入力し、最適ＡＣＢゲイン修正回路１４８０から出力される修正ＡＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。
【０１３３】
ＡＣＢゲイン符号化回路１４１０は、ＡＣＢゲインコードブック１４１１から順次読み込まれるＡＣＢゲインと入力端子９７からの最適ＡＣＢゲインとから第１の自乗誤差を計算し、ＡＣＢゲインと修正ＡＣＢゲインとから第２の自乗誤差を計算し、音声判定値から計算される重み係数と、第１の自乗誤差と、第２の自乗誤差とから次式で定義される評価関数を計算する。

【０１３４】
ここで、

は最適ＡＣＢゲイン、

は修正ＡＣＢゲイン、

はＡＣＢゲインコードブックから順次読み込まれるＡＣＢゲインであり、μは重み係数である。例えば、音声判定値Ｖｓが１（音声区間）のとき、重み係数μは１．０とし、Ｖｓが０（非音声区間）のときはμは０．２とする。
【０１３５】
そして、ＡＣＢゲイン符号化回路１４１０は、評価関数が最小となるＡＣＢゲインを選択し、選択されたＡＣＢゲインを第２のＡＣＢゲインとして第２の目標信号計算回路１４３０へ出力するとともに、第２の励振信号計算回路１６１０へ出力端子９５を介して出力し、第２のＡＣＢゲインに対応する符号をＡＣＢゲイン符号としてゲイン符号多重化回路１４７０へ出力する。
【０１３６】
最適ＦＣＢゲイン修正回路１４５０は、最適ＦＣＢゲイン計算回路１４４０から出力される最適ＦＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値Ｖｓを入力する。
【０１３７】
最適ＦＣＢゲイン修正回路１４５０において、音声判定値Ｖｓが０（非音声区間）のとき、最適ＦＣＢゲインの長時間平均を修正ＦＣＢゲインとする。非音声区間において、次式により最適ＦＣＢゲインの長時間平均を計算する。

【０１３８】
ここで、

は第ｎサブフレームにおける最適ＦＣＢゲイン、

は第ｎサブフレームにおける最適ＦＣＢゲインの長時間平均であり、αは例えば０．９である。なお、長時間平均には、平均値、中央値、最頻値なども適用できる。
【０１３９】
一方、最適ＦＣＢゲイン修正回路１４５０において、音声判定値Ｖｓが１（音声区間）のとき、最適ＦＣＢゲインそのものを修正ＦＣＢゲインとする。
【０１４０】
最適ＦＣＢゲイン修正回路１４５０は、修正ＦＣＢゲインをＦＣＢゲイン符号化回路１４２０へ出力する。
【０１４１】
ＦＣＢゲイン符号化回路１４２０は、最適ＦＣＢゲイン計算回路１４４０から出力される最適ＦＣＢゲインを入力し、最適ＦＣＢゲイン修正回路１４５０から出力される修正ＦＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。ＦＣＢゲイン符号化回路１４２０は、ＦＣＢゲインコードブック１４２１から順次読み込まれるＦＣＢゲインと、最適ＦＣＢゲインとから第１の自乗誤差を計算し、ＦＣＢゲインと修正ＦＣＢゲインとから第２の自乗誤差を計算し、音声判定値から計算される重み係数と第１の自乗誤差と第２の自乗誤差とから次式で定義される評価関数を計算する。

【０１４２】
ここで、

は最適ＦＣＢゲイン、

は修正ＦＣＢゲイン、

はＦＣＢゲインコードブックから順次読み込まれるＦＣＢゲインであり、μは重み係数である。例えば、音声判定値Ｖｓが１（音声区間）のとき、重み係数μは１．０とし、音声判定値Ｖｓが０（非音声区間）のときはμは０．２とする。
【０１４３】
そして、ＦＣＢゲイン符号化回路１４２０は、評価関数が最小となるＦＣＢゲインを選択し、選択されたＦＣＢゲインを第２のＦＣＢゲインとして第２の励振信号計算回路１６１０へ出力端子９６を介して出力し、第２のＦＣＢゲインに対応する符号をＦＣＢゲイン符号としてゲイン符号多重化回路１４７０へ出力する。
【０１４４】
ゲイン符号多重回路１４７０は、ＡＣＢゲイン符号化回路１４１０から出力されるＡＣＢゲイン符号を入力し、ＦＣＢゲイン符号化回路１４２０から出力されるＦＣＢゲイン符号を入力し、ＡＣＢゲイン符号とＦＣＢゲイン符号とを多重化して得られる第２のゲイン符号を、方式Ｂにおけるゲイン復号方法により復号可能な符号として符号多重回路１０２０へ出力端子５６を介して出力する。
【０１４５】
以上で図８によるゲイン符号生成回路１４００の説明を終え、再び図１の説明に戻る。
【０１４６】
第２の励振信号計算回路１６１０は、目標信号計算回路１７００から出力される第２のＡＣＢ信号を入力し、ＦＣＢ符号生成回路１８００から出力される第２のＦＣＢ信号を入力し、ゲイン符号生成回路１４００から出力される第２のＡＣＢゲインと第２のＦＣＢゲインとを入力する。第２の励振信号計算回路１６１０は、第２のＡＣＢ信号に第２のＡＣＢゲインを乗じて得た信号と、第２のＦＣＢ信号に第２のＦＣＢゲインを乗じて得た信号と、を加算して第２の励振信号を得る。そして第２の励振信号を第２の励振信号記憶回路１６２０へ出力する。
【０１４７】
第２の励振信号記憶回路１６２０は、第２の励振信号計算回路１６１０から出力される第２の励振信号を入力し、これを記憶保持する。そして、過去に入力されて記憶保持されている第２の励振信号を目標信号計算回路１７００へ出力する。以上により、本発明の第１の実施例の説明を終える。
【０１４８】
次に、本発明の第２の実施例について説明する。図９は、本発明による符号変換装置の第２の実施例の構成を示す図である。図９においては、図１２におけるＬＰ係数符号変換回路１００と、ゲイン符号変換回路４００とを、それぞれＬＰ係数符号変換回路１１００とゲイン符号変換回路２４００とで置き換え、ＬＰ係数符号変換回路１１００とゲイン符号変換回路２４００との間に結線が付加されている。以下では、図１２に示す要素と同一又は同等の要素の説明は省略し、相違点について説明する。
【０１４９】
ＬＰ係数符号変換回路１１００は、図１を用いて説明した第１の実施例におけるそれと同様である。ただし、他回路との結線の仕方が異なっており、第１のＬＳＰをゲイン符号変換回路４００へ出力する。
【０１５０】
ゲイン符号変換回路２４００は、符号分離回路１０１０から出力される第１のゲイン符号を入力し、ＬＰ係数符号変換回路１１００から出力される第１のＬＳＰを入力する。
【０１５１】
ゲイン符号変換回路２４００は、まず、第１のゲイン符号を、方式Ａにおけるゲイン復号方法により復号して得られる第１のゲイン（第１のＡＣＢゲイン及び第１のＦＣＢゲイン）から、修正ＡＣＢゲイン及び修正ＦＣＢゲインを計算し、第１のＬＳＰから音声判定値を決定する。
【０１５２】
次に、ゲイン符号変換回路２４００は、ＡＣＢゲインコードブックから順次読み込まれるＡＣＢゲインと第１のＡＣＢゲインとから第１の自乗誤差を計算し、ＡＣＢゲインと修正ＡＣＢゲインとから第２の自乗誤差を計算する。
【０１５３】
そして、ゲイン符号変換回路２４００は、音声判定値から計算される重み係数と、第１の自乗誤差と、第２の自乗誤差とから計算される評価関数が最小となるＡＣＢゲイン及び対応するＡＣＢゲイン符号を選択する。
【０１５４】
また、ゲイン符号変換回路２４００は、ＦＣＢゲインコードブックから順次読み込まれるＦＣＢゲインと第１のＦＣＢゲインとから第３の自乗誤差を計算し、ＦＣＢゲインと修正ＦＣＢゲインとから第４の自乗誤差を計算する。そして、ゲイン符号変換回路２４００は、音声判定値から計算される重み係数と第３の自乗誤差と第４の自乗誤差とから計算される評価関数が最小となるＦＣＢゲイン及び対応するＦＣＢゲイン符号を選択する。
【０１５５】
最後に、ゲイン符号変換回路２４００は、選択されたＡＣＢゲイン符号とＦＣＢゲイン符号とからなる第２のゲイン符号を、方式Ｂにおけるゲイン復号方法により復号可能な符号として符号多重回路１０２０へ出力する。
【０１５６】
図１０は、図９のゲイン符号変換回路２４００の構成を示す図である。図１０を参照すると、ゲイン符号変換回路２４００は、音声／非音声識別回路１４６０と、ゲイン符号分離回路２４９０と、ＡＣＢゲイン復号回路２４７０と、ＡＣＢゲインコードブック２４７１と、ＡＣＢゲイン修正回路２４４０と、ＡＣＢゲイン符号化回路２４１０と、ＡＣＢゲインコードブック１４１１と、ＦＣＢゲイン復号回路２４８０と、ＦＣＢゲインコードブック２４８１と、ＦＣＢゲイン修正回路２４５０と、ＦＣＢゲイン符号化回路２４２０と、ＦＣＢゲインコードブック１４２１と、ゲイン符号多重回路１４７０と、を備えている。図１０を参照して、この実施例のゲイン符号変換回路２４００の各構成要素について説明する。なお、図１０において、音声／非音声識別回路１４６０及びゲイン符号多重回路１４７０は、図８に示した要素と基本的に同じであり、以下では、これらの説明は省略する。
【０１５７】
ゲイン符号分離回路２４９０は、符号分離回路１０１０から出力される第１のゲイン符号を入力端子４５を介して入力し、第１のゲイン符号からＡＣＢゲイン及びＦＣＢゲインに対応する符号、すなわち第１のＡＣＢゲイン符号及び第１のＦＣＢゲイン符号を分離し、第１のＡＣＢゲイン符号をＡＣＢゲイン復号回路２４７０へ出力し、第１のＦＣＢゲイン符号をＦＣＢゲイン復号回路２４８０へ出力する。
【０１５８】
ＡＣＢゲイン復号回路２４７０は、複数セットのＡＣＢゲインが格納されたＡＣＢゲインコードブック２４７１を備えており、ゲイン符号分離回路２４９０から出力される第１のＡＣＢゲイン符号を入力し、第１のＡＣＢゲイン符号に対応するＡＣＢゲインを第１のＡＣＢゲインコードブック２４７１より読み出し、読み出されたＡＣＢゲインを第１のＡＣＢゲインとしてＡＣＢゲイン修正回路２４４０へ出力するとともに、ＡＣＢゲイン符号化回路２４１０へ出力する。ここで、ＡＣＢゲイン符号からのＡＣＢゲインの復号は、方式ＡにおけるＡＣＢゲインの復号方法に従い、方式ＡのＡＣＢゲインコードブックを用いる。
【０１５９】
ＦＣＢゲイン復号回路２４８０は、複数セットのＦＣＢゲインが格納されたＦＣＢゲインコードブック２４８１を備えており、ゲイン符号分離回路２４９０から出力される第１のＦＣＢゲイン符号を入力し、第１のＦＣＢゲイン符号に対応するＦＣＢゲインを第１のＦＣＢゲインコードブック２４８１より読み出し、読み出されたＦＣＢゲインを第１のＦＣＢゲインとしてＦＣＢゲイン修正回路２４５０へ出力するとともに、ＦＣＢゲイン符号化回路２４２０へ出力する。ここで、ＦＣＢゲイン符号からのＦＣＢゲインの復号は、方式ＡにおけるＦＣＢゲインの復号方法に従い、方式ＡのＦＣＢゲインコードブックを用いる。
【０１６０】
ＡＣＢゲイン修正回路２４４０は、ＡＣＢゲイン復号回路２４７０から出力される第１のＡＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。音声判定値Ｖｓが０（非音声区間）のとき、第１のＡＣＢゲインの長時間平均を修正ＡＣＢゲインとする。
【０１６１】
ＡＣＢゲイン修正回路２４４０は、非音声区間において、次式により第１のＡＣＢゲインの長時間平均を計算する。

【０１６２】
ここで、

は第ｎサブフレームにおける第１のＡＣＢゲイン、

は第ｎサブフレームにおける第１のＡＣＢゲインの長時間平均であり、αは例えば０．９である。なお、長時間平均には、平均値、中央値、最頻値なども適用できる。
【０１６３】
一方、音声判定値Ｖｓが１（音声区間）のとき、ＡＣＢゲイン修正回路２４４０は、第１のＡＣＢゲインそのものを修正ＡＣＢゲインとする。
【０１６４】
ＡＣＢゲイン修正回路２４４０は、修正ＡＣＢゲインをＡＣＢゲイン符号化回路２４１０へ出力する。
【０１６５】
ＦＣＢゲイン修正回路２４５０は、ＦＣＢゲイン復号回路２４８０から出力される第１のＦＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。
【０１６６】
ＦＣＢゲイン修正回路２４５０において、音声判定値Ｖｓが０（非音声区間）のとき、第１のＦＣＢゲインの長時間平均を修正ＦＣＢゲインとする。非音声区間において、次式により第１のＦＣＢゲインの長時間平均を計算する。

【０１６７】
ここで、

は第ｎサブフレームにおける第１のＦＣＢゲイン、

は第ｎサブフレームにおける第１のＦＣＢゲインの長時間平均であり、αは例えば０．９である。なお、長時間平均には、平均値、中央値、最頻値なども適用できる。
【０１６８】
一方、音声判定値Ｖｓが１（音声区間）のとき、ＦＣＢゲイン修正回路２４５０は、第１のＦＣＢゲインそのものを修正ＦＣＢゲインとする。
【０１６９】
ＦＣＢゲイン修正回路２４５０は、修正ＦＣＢゲインをＦＣＢゲイン符号化回路２４２０へ出力する。
【０１７０】
ＡＣＢゲイン符号化回路２４１０は、ＡＣＢゲイン復号回路２４７０から出力される第１のＡＣＢゲインを入力し、ＡＣＢゲイン修正回路２４４０から出力される修正ＡＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。
【０１７１】
ＡＣＢゲイン符号化回路２４１０は、ＡＣＢゲインコードブック１４１１から順次読み込まれるＡＣＢゲインと第１のＡＣＢゲインとから第１の自乗誤差を計算し、ＡＣＢゲインと修正ＡＣＢゲインとから第２の自乗誤差を計算し、音声判定値から計算される重み係数と第１の自乗誤差と第２の自乗誤差とから次式で定義される評価関数を計算する。
【０１７２】

【０１７３】
ここで、

は第１のＡＣＢゲイン、

は修正ＡＣＢゲイン、

はＡＣＢゲインコードブック１４１１から順次読み込まれるＡＣＢゲインであり、μは重み係数である。例えば、音声判定値Ｖｓが１（音声区間）のとき、重み係数μは１．０とし、Ｖｓが０（非音声区間）のときはμは０．２とする。
【０１７４】
そして、ＡＣＢゲイン符号化回路２４１０は、評価関数が最小となるＡＣＢゲインを選択し、選択されたＡＣＢゲインを第２のＡＣＢゲインとし、第２のＡＣＢゲインに対応する符号を第２のＡＣＢゲイン符号としてゲイン符号多重化回路１４７０へ出力する。
【０１７５】
ＦＣＢゲイン符号化回路２４２０は、ＦＣＢゲイン復号回路２４８０から出力される第１のＦＣＢゲインを入力し、ＦＣＢゲイン修正回路２４５０から出力される修正ＦＣＢゲインを入力し、音声／非音声識別回路１４６０から出力される音声判定値を入力する。
【０１７６】
ＦＣＢゲイン符号化回路２４２０は、ＦＣＢゲインコードブック１４２１から順次読み込まれるＦＣＢゲインと第１のＦＣＢゲインとから第３の自乗誤差を計算し、ＦＣＢゲインと修正ＦＣＢゲインとから第４の自乗誤差を計算し、音声判定値から計算される重み係数と第３の自乗誤差と第４の自乗誤差とから次式で定義される評価関数を計算する。

【０１７７】
ここで、

は第１のＦＣＢゲイン、

は修正ＦＣＢゲイン、

はＦＣＢゲインコードブック１４２１から順次読み込まれるＦＣＢゲインであり、μは重み係数である。例えば、音声判定値Ｖｓが１（音声区間）のとき、重み係数μは１．０とし、音声判定値Ｖｓが０（非音声区間）のときはμは０．２とする。
【０１７８】
そして、ＦＣＢゲイン符号化回路２４２０は、評価関数が最小となるＦＣＢゲインを選択し、選択されたＦＣＢゲインを第２のＦＣＢゲインとし、第２のＦＣＢゲインに対応する符号を第２のＦＣＢゲイン符号としてゲイン符号多重化回路１４７０へ出力する。
【０１７９】
上述した本発明の各実施例の符号変換装置は、ディジタル信号処理プロセッサ等のコンピュータ制御で実現するようにしてもよい。図１１は本発明の第３の実施例として、上記各実施例の符号変換処理をコンピュータで実現する場合の装置構成を模式的に示す図である。記録媒体６から読み出されたプログラムを実行するコンピュータ１において、第１の符号化復号装置により音声を符号化して得た第１の符号を第２の符号化復号装置により復号可能な第２の符号へ変換する符号変換処理を実行するにあたり、記録媒体６には、
（ａ）　第１の符号列から第１の線形予測係数を得る処理と、
（ｂ）　第１の符号列から励振信号の情報を得る処理と、
（ｃ）　励振信号の情報から励振信号を得る処理と、
（ｄ）　第１の線形予測係数をもつフィルタを励振信号により駆動することによって音声信号を生成する処理と、
（ｅ）　第２の符号列から得られる情報により生成される第２の音声信号と、第１の音声信号との距離が最小となるゲイン（最適ゲイン）を計算する処理と、
（ｆ）　最適ゲインを修正する処理と、
（ｇ）　修正された最適ゲイン（修正最適ゲイン）と、第２の方式におけるゲインコードブックから読み出されるゲインとから第１の自乗誤差を計算し、最適ゲインと、ゲインコードブックから読み出されるゲインとから第２の自乗誤差を計算し、第１の自乗誤差と第２の自乗誤差に基づく評価関数が最小となるゲインをゲインコードブックから選択することによって第２の符号列におけるゲイン情報を求める処理、
を実行させるためのプログラムが記録されている。記録媒体６から該プログラムを記録媒体読出装置５、インタフェース４を介してメモリ３に読み出して実行する。上記プログラムは、マスクＲＯＭ等、フラッシュメモリ等の不揮発性メモリに格納してもよく、記録媒体は不揮発性メモリを含むほか、ＣＤ−ＲＯＭ、ＦＤ、Ｄｉｇｉｔａｌ　Ｖｅｒｓａｔｉｌｅ　Ｄｉｓｋ　（ＤＶＤ）、磁気テープ（ＭＴ）、可搬型ＨＤＤ等の媒体の他、例えばサーバ装置からコンピュータで該プログラムを通信媒体伝送する場合等、プログラムを担持する有線、無線で通信される通信媒体等も含む。
【０１８０】
本発明の第４の実施例では、記録媒体６から読み出されたプログラムを実行するコンピュータ１において、第１の符号化復号装置により音声を符号化して得た第１の符号を第２の符号化復号装置により復号可能な第２の符号へ変換する符号変換処理を実行するにあたり、記録媒体６には、
（ａ）　第１の符号列からゲイン情報を復号する処理と、
（ｂ）　復号されたゲイン（復号ゲイン）を修正する処理と、
（ｃ）　修正された復号ゲイン（修正復号ゲイン）と、第２の方式におけるゲインコードブックから読み出されるゲインとから第１の自乗誤差を計算し、復号ゲインと、ゲインコードブックから読み出されるゲインとから第２の自乗誤差を計算し、第１の自乗誤差と第２の自乗誤差に基づく評価関数が最小となるゲインをゲインコードブックから選択することによって第２の符号列におけるゲイン情報を求める処理、
を実行させるためのプログラムが記録されている。
【０１８１】
以上本発明を上記実施例に即して説明したが、本発明は、上記実施例の構成にのみ限定されるものでなく、特許請求の範囲の各請求項の発明の範囲内で当業者であればなし得るであろう各種変形、修正を含むことは勿論である。
【０１８２】
【発明の効果】
以上説明したように、本発明によれば、非音声区間における背景雑音音質の劣化を低減することができる、という効果を奏する。
【０１８３】
その理由は、本発明においては、第１の符号列から第１の線形予測係数をもつ合成フィルタを励振信号で駆動して得た第１の音声信号と第２の符号列から得られる情報により生成される第２の音声信号とから最適ゲインを導出し、さらに最適ゲインを修正し、修正した最適ゲインと、最適ゲインと、第２の方式におけるゲインコードブックから読み出されるゲインとに基づき、第２の符号列におけるゲイン情報を求め、その際、非音声区間において、第２のゲインの時間変動が小さくなるような評価関数を用いて、第２のゲインを求めるように構成したためである。上記効果は、第１の符号列からゲイン情報を復号し、復号されたゲインを修正し、修正された復号ゲインと、前記復号ゲインと第２の方式におけるゲインコードブックから読み出されるゲインとに基づき、第２の符号列におけるゲイン情報を求め、非音声区間において、第２のゲインの時間変動が小さくなるような評価関数を用いて、第２のゲインを求めるように構成してなる本発明によっても奏することができる。
【図面の簡単な説明】
【図１】本発明による符号変換装置の第１の実施例の構成を示す図である。
【図２】本発明による符号変換装置におけるＬＰ係数符号変換回路の構成を示す図である。
【図３】ＡＣＢ符号とＡＣＢ遅延との対応関係とＡＣＢ符号の読み替え方法を説明する図である。
【図４】本発明による符号変換装置の音声復号回路の構成を示す図である。
【図５】本発明による符号変換装置における目標信号計算回路の構成を示す図である。
【図６】本発明による符号変換装置におけるＦＣＢ符号生成回路の構成を示す図である。
【図７】パルス位置符号とパルス位置との対応関係とＡＣＢ符号の読み替え方法を説明する図である。
【図８】本発明による符号変換装置におけるゲイン符号生成回路の構成を示す図である。
【図９】本発明による符号変換装置の第２の実施例の構成を示す図である。
【図１０】本発明による符号変換装置におけるゲイン符号生成回路の構成を示す図である。
【図１１】本発明による符号変換装置の第３から第４の実施例の構成を示す図である。
【図１２】従来の符号変換装置の構成を示す図である。
【符号の説明】
１　コンピュータ
２　ＣＰＵ
３　メモリ
４　記録媒体読出装置インタフェース
５　記録媒体読出装置
６　記録媒体
１０，３１，３５，３６，３７，５１，５２，５３，５７，６１，７４，７５，８１，８２，８３，８４，８５，９１，９２，９３，９４　入力端子
２０，３２，３３，３４，５５，５６，６２，６３，７６，７７，７８，８６，９５，９６　出力端子
１００，１１００　ＬＰ係数符号変換回路
１１０　ＬＰ係数復号回路
１３０　ＬＰ係数符号化回路
１１１　第１のＬＳＰコードブック
１３１　第２のＬＳＰコードブック
２００，１２００　ＡＣＢ　符号変換回路
３００，１３００　ＦＣＢ　符号変換回路
４００，２４００　ゲイン符号変換回路
１０１０　符号分離回路
１０２０　符号多重回路
１１１０　ＬＳＰ−ＬＰＣ変換回路
１１２０　インパルス応答計算回路
１４００　ゲイン符号生成回路
１４１０，２４１０　ＡＣＢゲイン符号化回路
１４１１，２４７１　ＡＣＢゲインコードブック
１４２０，２４２０　ＦＣＢゲイン符号化回路
１４２１，２４８１　ＦＣＢゲインコードブック
１４３０　第２の目標信号計算回路
１４４０　最適ＦＣＢゲイン計算回路
１４５０　最適ＦＣＢゲイン修正回路
１４６０　音声／非音声識別回路
１４７０　ゲイン符号多重回路
１４８０　最適ＡＣＢゲイン修正回路
１５００　音声復号回路
１５１０　ＡＣＢ復号回路
１５２０　ＦＣＢ復号回路
１５３０　ゲイン復号回路
１５４０　励振信号計算回路
１５７０　励振信号記憶回路
１５８０　合成フィルタ
１６００　励振信号情報復号回路
１６１０　第２の励振信号計算回路
１６２０　第２の励振信号記憶回路
１７００　目標信号計算回路
１７１０　重み付け信号計算回路
１７２０　ＡＣＢ信号生成回路
１８００　ＦＣＢ符号生成回路
１８２０　ＦＣＢ信号生成回路
２４８０　ＦＣＢゲイン復号回路
２４５０　ＦＣＢゲイン修正回路
２４９０　ゲイン符号分離回路[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an encoding and decoding method for transmitting or storing an audio signal at a low bit rate, and in particular, when performing audio communication using different encoding and decoding methods, obtained by encoding audio by a certain method. The present invention relates to a code conversion method and apparatus for converting a code into a code that can be decoded by another method with high sound quality and a low operation amount, and a recording medium thereof.
[0002]
[Prior art]
2. Description of the Related Art As a method for encoding a speech signal at a medium to low bit rate with high efficiency, a method of separating and encoding the speech signal into a linear prediction (Linear Prediction: LP) filter and an excitation signal for driving the filter is widely used. . One of the typical methods is Code \ Excited \ Linear \ Prediction (code-excited linear prediction: referred to as "CELP"). In CELP, an LP filter in which LP coefficients representing the frequency characteristics of an input voice are set includes an adaptive codebook (Adaptive Codebook: \ "ACB") representing a pitch period of the input voice, and a fixed codebook (Random Number and Pulse) composed of random numbers and pulses. By driving with an excitation signal represented by the sum of “Fixed Codebook:“ FCB ”), a synthesized speech signal is obtained. At this time, the ACB component and the FCB component are multiplied by gains (“ACB gain” and “FCB gain”). For CELP, M.P. Schroeder and B.S. S. "Code \ excited \ linear \ prediction: \ High \ quality \ speech \ at \ very \ low \ bit \ rates" by Atal (Proc. \ Of \ IEEE \ Int. \ Conf.on \ Acost., @Spence. Is referred to.
[0003]
By the way, for example, when an interconnection between a 3G mobile network and a wired packet network is assumed, there is a problem that a direct connection cannot be made because the standard voice coding method used in each network is different. The simplest solution to this is a tandem connection. However, in the tandem connection, the audio signal is temporarily decoded using the standard system from a code string obtained by encoding the audio using one standard system, and the decoded audio signal is decoded using the other standard system. The encoding is performed again. For this reason, there is a problem that the sound quality is generally lowered, the delay is increased, and the calculation amount is increased as compared with the case where encoding and decoding are performed only once in each audio encoding / decoding method.
[0004]
On the other hand, a code obtained by encoding speech using one standard method is converted into a code that can be decoded by the other standard method in a code domain or a coding parameter domain. Is effective for The method of converting the code is described in "Improving, Transcoding, Capability, of Speech, Coders, in Clean, and Frame, Erased, Channel, Environmental, Environmental, Environmental Engineering, Hong Kong, Kong et al." Reference 2)).
[0005]
FIG. 12 converts a code obtained by encoding speech using the first speech coding method (referred to as “method A”) into a code that can be decoded using the second method (referred to as “method B”). FIG. 3 is a diagram illustrating an example of a configuration of a transcoder. Referring to FIG. 12, the code conversion device includes an input terminal 10, a code separation circuit 1010, an LP coefficient code conversion circuit 100, an ACB code conversion circuit 200, an FCB code conversion circuit 300, and a gain code conversion circuit 400. , A code multiplexing circuit 1020, and an output terminal 20. With reference to FIG. 12, each component of the conventional transcoder will be described.
[0006]
From the input terminal 10, a first code string obtained by encoding a sound by the method A is input.
[0007]
The code separation circuit 1010 converts, from the first code string input from the input terminal 10, codes corresponding to the LP coefficient, ACB, FCB, ACB gain, and FCB gain, that is, the LP coefficient code, ACB code, FCB code, and gain code. To separate. Here, the ACB gain and the FCB gain are collectively encoded and decoded, and for simplicity, this is referred to as a gain and its code is referred to as a gain code. Further, the LP coefficient code, the ACB code, the FCB code, and the gain code are referred to as a first LP coefficient code, a first ACB code, a first FCB code, and a first gain code, respectively. Then, the first LP coefficient code is output to the LP coefficient code conversion circuit 100, the first ACB code is output to the ACB code conversion circuit 200, and the first FCB code is output to the FCB code conversion circuit 300. The gain code of 1 is output to the gain code conversion circuit 400.
[0008]
The LP coefficient code conversion circuit 100 receives the first LP coefficient code output from the code separation circuit 1010, and converts the first LP coefficient code into a code that can be decoded by the method B. The converted LP coefficient code is output to code multiplexing circuit 1020 as a second LP coefficient code.
[0009]
The ACB code conversion circuit 200 receives the first ACB code output from the code separation circuit 1010 and converts the first ACB code into a code that can be decoded by the method B. The converted ACB code is output to code multiplexing circuit 1020 as a second ACB code.
[0010]
The FCB code conversion circuit 300 receives the first FCB code output from the code separation circuit 1010, and converts the first FCB code into a code that can be decoded by the method B. The converted FCB code is output to code multiplexing circuit 1020 as a second FCB code.
[0011]
The gain code conversion circuit 400 receives the first gain code output from the code separation circuit 1010 and converts the first gain code into a code that can be decoded by the method B. The converted gain code is output to code multiplexing circuit 1020 as a second gain code.
[0012]
A more specific operation of each conversion circuit will be described below.
[0013]
The LP coefficient code conversion circuit 100 decodes the first LP coefficient code input from the code separation circuit 1010 by the LP coefficient decoding method in the system A to obtain a first LP coefficient. Next, the LP coefficient code conversion circuit 100 quantizes and encodes the first LP coefficient according to the method for quantizing and encoding the LP coefficient in the method B to obtain a second LP coefficient code. Then, LP coefficient code conversion circuit 100 outputs the second LP coefficient code to code multiplexing circuit 1020 as a code that can be decoded by the LP coefficient decoding method in scheme B.
[0014]
The ACB code conversion circuit 200 obtains a second ACB code by reading the first ACB code input from the code separation circuit 1010 using the correspondence between the code in the method A and the code in the method B. Then, ACB code conversion circuit 200 outputs the second ACB code to code multiplexing circuit 1020 as a code that can be decoded by the ACB decoding method in scheme B.
[0015]
The FCB code conversion circuit 300 obtains a second FCB code by reading the first FCB code input from the code separation circuit 1010 using the correspondence between the code in the method A and the code in the method B. Then, FCB code conversion circuit 300 outputs the second FCB code to code multiplexing circuit 1020 as a code that can be decoded by the FCB decoding method in scheme B.
[0016]
The gain code conversion circuit 400 obtains a first gain by decoding the first gain code input from the code separation circuit 1010 by the gain decoding method in the scheme A. Next, the gain code conversion circuit 400 quantizes and codes the first gain by the gain quantization method and the coding method in the method B, and obtains a second gain and its sign (second gain code). Get. Then, gain code conversion circuit 400 outputs the second gain code to code multiplexing circuit 1020 as a code that can be decoded by the gain decoding method in scheme B.
[0017]
The code multiplexing circuit 1020 includes a second LP coefficient code output from the LP coefficient code conversion circuit 100, a second ACB code output from the ACB code conversion circuit 200, and a second LP coefficient code output from the FCB code conversion circuit 300. The second FCB code and the second gain code output from the gain code conversion circuit 400 are input, and a code string obtained by multiplexing them is output as a second code string via the output terminal 20. Thus, the description of FIG. 12 is completed.
[0018]
[Problems to be solved by the invention]
However, the conventional transcoder described with reference to FIG. 12 has a problem that the sound quality of background noise in a non-speech section deteriorates.
[0019]
The reason is that the background noise energy has a large time variation in the non-voice section. This is due to the fact that the second gain obtained by requantizing the first gain fluctuates greatly in a non-voice section.
[0020]
Accordingly, the present invention has been made in view of the above problems, and a main object of the present invention is to provide an apparatus and a method capable of reducing deterioration of background noise sound quality in a non-speech section, and a recording medium recording the program thereof. It is in. Other objects, features, advantages and the like of the present invention will be immediately apparent to those skilled in the art from the following description.
[0021]
[Means for Solving the Problems]
To achieve the above object, a method according to a first aspect of the present invention is a code conversion method for converting a first code string conforming to a first method into a second code string conforming to a second method. , By obtaining information of a first linear prediction coefficient and an excitation signal from the first code string, and driving a filter having the first linear prediction coefficient with an excitation signal obtained from the information of the excitation signal. Generating a first audio signal, calculating an optimal gain based on a second audio signal generated based on information obtained from a second code sequence, and the first audio signal; Correcting the gain in the second code string based on the step of correcting the gain, the corrected optimum gain (corrected optimum gain), the optimum gain, and the gain read from the gain codebook in the second method. Comprising a step of obtaining information, the. In the method according to the present invention, the optimum gain is preferably obtained as a gain that minimizes a distance between a second audio signal generated based on information obtained from a second code string and the first audio signal. Can be
[0022]
A method according to a second aspect of the present invention is a code conversion method for converting a first code string conforming to a first method into a second code string conforming to a second method, A step of decoding gain information from a code string, a step of correcting a decoded gain (decoding gain), a corrected decoding gain (corrected decoding gain), the decoding gain, and a gain codebook in a second method. And obtaining gain information in the second code string based on the gain read from.
[0023]
In the invention according to the first aspect, preferably, a first squared error is calculated from the corrected optimal gain and a gain read from the gain codebook, and the first squared error is read from the optimal gain and the gain codebook. A second square error is calculated from the gain and the gain that minimizes the evaluation function based on the first square error and the second square error is selected from the gain codebook. Find gain information.
[0024]
In the invention according to the second aspect, preferably, a first squared error is calculated from the corrected decoding gain and a gain read from the gain codebook, and the first squared error is read from the decoding gain and the gain codebook. A second square error is calculated from the gain and the gain that minimizes the evaluation function based on the first square error and the second square error is selected from the gain codebook. Find gain information.
[0025]
In the invention according to the first aspect, preferably, the corrected optimum gain is based on a long-term average of the optimum gain.
[0026]
In the invention according to the second aspect, preferably, the modified decoding gain is based on a long-term average of the decoding gain.
[0027]
An apparatus according to a third aspect of the present invention is directed to a code conversion apparatus for converting a first code string conforming to a first method into a second code string conforming to a second method, A first audio signal is obtained by obtaining information of a first linear prediction coefficient and an excitation signal from a code string, and driving a filter having the first linear prediction coefficient with an excitation signal obtained from the information of the excitation signal. An audio decoding circuit for generating, an optimal gain calculation circuit for calculating an optimal gain based on a second audio signal generated based on information obtained from a second code string, and the first audio signal; An optimum gain correction circuit for correcting the gain, a corrected optimum gain (corrected optimum gain), the optimum gain, and a gain read out from a gain codebook in the second method; Gain encoding circuit for finding the gain information, including. In the apparatus according to the present invention, the optimum gain calculation circuit preferably includes a gain that minimizes a distance between a second audio signal generated by information obtained from a second code string and the first audio signal. Is determined as the optimum gain.
[0028]
An apparatus according to a fourth aspect of the present invention is the code conversion apparatus for converting a first code string conforming to a first method into a second code string conforming to a second method, A gain decoding circuit for decoding gain information from the code string, a decoding gain correction circuit for correcting the decoded gain (decoding gain), a corrected decoding gain (corrected decoding gain), the decoding gain, A gain coding circuit for obtaining gain information in the second code sequence based on the gain read from the gain codebook in the method.
[0029]
In the invention according to the third aspect, the gain encoding circuit preferably calculates a first squared error from the corrected optimal gain and a gain read from the gain codebook, and calculates the optimal gain, Calculating a second squared error from the gain read out from the gain codebook, and selecting from the gain codebook a gain that minimizes the evaluation function based on the first squared error and the second squared error. Gain information in the second code string is obtained.
[0030]
In the invention according to the fourth aspect, the gain encoding circuit preferably calculates a first square error from the corrected decoding gain and a gain read from the gain codebook, and calculates the decoding gain, Calculating a second squared error from the gain read out from the gain codebook, and selecting from the gain codebook a gain that minimizes the evaluation function based on the first squared error and the second squared error. Gain information in the second code string is obtained.
[0031]
In the above-described optimal gain correction circuit according to the third aspect, preferably, the corrected optimal gain is based on a long-term average of the optimal gain.
[0032]
In the decoding gain correction circuit according to the fourth aspect, preferably, the correction decoding gain is based on a long-term average of the decoding gain.
[0033]
According to a fifth aspect of the present invention, there is provided a program configured to convert a first code string conforming to the first method into a second code string conforming to the second method.
(A) Obtaining information of a first linear prediction coefficient and an excitation signal from the first code string, and driving a filter having the first linear prediction coefficient with an excitation signal obtained from the information of the excitation signal. Generating a first audio signal by
(B) a process of calculating a gain (optimum gain) based on a second audio signal generated based on information obtained from a second code sequence and the first audio signal;
(C) processing for correcting the optimum gain;
(D) executing a process of obtaining gain information in the second code sequence based on the corrected optimum gain (corrected optimum gain), the optimum gain, and a gain read from a gain codebook in the second method. Provide a program to make In the present invention, the gain at which the distance between the second audio signal generated from the information obtained from the second code string and the first audio signal is minimized is determined as the optimum gain.
[0034]
A program according to a sixth aspect of the present invention provides a computer that constitutes a code conversion device that converts a first code string conforming to the first method into a second code string conforming to the second method.
(A) decoding gain information from the first code string;
(B) processing for correcting the decoded gain (decoding gain);
(C) performing a process of obtaining gain information in a second code string based on a corrected decoding gain (corrected decoding gain), the decoding gain, and a gain read from a gain codebook in the second method. Provide a program to make
[0035]
In the program according to the fifth aspect, preferably, a first squared error is calculated from the corrected optimum gain and a gain read from the gain codebook, and the first squared error is calculated from the optimum gain and the gain codebook. The second code is calculated by calculating a second square error from the read gain and selecting a gain from the gain codebook that minimizes an evaluation function based on the first square error and the second square error. Find the gain information in a column.
[0036]
In the program of the invention according to the sixth aspect, preferably, a first square error is calculated from the corrected decoding gain and a gain read from the gain codebook, and the first squared error is calculated from the decoding gain and the gain codebook. The second code is calculated by calculating a second square error from the read gain and selecting a gain from the gain codebook that minimizes an evaluation function based on the first square error and the second square error. Find the gain information in a column.
[0037]
In the program according to the fifth aspect, preferably, the corrected optimal gain is based on a long-term average of the optimal gain.
[0038]
In the program according to the sixth aspect, preferably, the modified decoding gain is based on a long-term average of the decoding gain.
[0039]
The invention according to a seventh aspect of the present invention provides a recording medium on which the program according to the fifth and sixth aspects is recorded.
[0040]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described. First, the outline and principle of the apparatus and method of the present invention will be described, and then embodiments will be described in detail below.
[0041]
In the code conversion device according to the present invention, the speech decoding circuit (1500) obtains information of a first linear prediction coefficient and an excitation signal from a first code string conforming to a first method, and A first audio signal is generated by driving a filter having a prediction coefficient with an excitation signal obtained from the information of the excitation signal, and a gain code generation circuit (1400) generates a second code conforming to a second method. Calculating a gain (optimum gain) that minimizes the distance between the second audio signal generated from the information obtained from the column and the first audio signal, modifying the optimal gain, and modifying the optimal gain Based on the (corrected optimum gain), the optimum gain, and the gain read from the gain codebook in the second method, gain information in the second code string is obtained.
[0042]
The method according to the present invention has the following steps.
[0043]
Step a: Obtain a first linear prediction coefficient from a first code string.
[0044]
Step b: Obtain information of the excitation signal from the first code string.
[0045]
Step c: An excitation signal is obtained from the information of the excitation signal.
[0046]
Step d: generating a first audio signal by driving a filter having a first linear prediction coefficient with the excitation signal.
[0047]
Step e: Calculate a gain (optimum gain) that minimizes the distance between the second audio signal generated based on the information obtained from the second code sequence and the first audio signal.
[0048]
Step f: Modify the optimum gain.
[0049]
Step g: Determine gain information in the second code sequence based on the corrected optimal gain (modified optimal gain), the optimal gain, and the gain read from the gain codebook in the second method.
[0050]
In the present invention, the second gain is obtained by using an evaluation function that reduces the time variation of the second gain in a non-voice section.
[0051]
For this reason, in the non-speech section, the time variation of the obtained second gain is small, and the time variation of the background noise energy in the section is small.
[0052]
As a result, it is possible to reduce the deterioration of the background noise sound quality in the non-voice section.
[0053]
【Example】
Next, embodiments of the present invention will be described in detail with reference to the drawings.
[0054]
FIG. 1 is a diagram showing a configuration of a first embodiment of a transcoder according to the present invention. 1, the same or equivalent elements as those in FIG. 12 are denoted by the same reference numerals. Referring to FIG. 1, input terminal 10, code separation circuit 1010, LP coefficient code conversion circuit 1100, LSP-LPC conversion circuit 1110, impulse response calculation circuit 1120, ACB code conversion circuit 1200, target signal calculation A circuit 1700, an FCB code generation circuit 1800, a gain code generation circuit 1400, a speech decoding circuit 1500, a second excitation signal calculation circuit 1610, a second excitation signal storage circuit 1620, a code multiplexing circuit 1020, And an output terminal 20. The input terminal 10, the output terminal 20, the code separation circuit 1010, and the code multiplexing circuit 1020 are basically the same as the elements shown in FIG. 12 except that a part of the connection is branched. In the following, description of the same or equivalent elements described above will be omitted, and differences from the configuration illustrated in FIG. 12 will be mainly described.
[0055]
Also, in scheme A, the encoding of the LP coefficient is

It is performed every msec cycle (frame), and encoding of components of the excitation signal such as ACB, FCB and gain is performed by:

It is performed every msec cycle (subframe).
[0056]
On the other hand, in scheme B, the encoding of the LP coefficient is

It is performed every msec cycle (frame), and the components of the excitation signal are encoded as follows:

It is performed every msec cycle (subframe).
[0057]
Further, the frame length, the number of subframes, and the subframe length of the method A are respectively

,

as well as

And
[0058]
The frame length, the number of subframes, and the subframe length of method B are

,

as well as,

And
[0059]
In the following description, for simplicity,

And
[0060]
Here, for example, the sampling frequency is 8000 Hz,

as well as

Is 10 msec,

as well as

Is 160 samples,

as well as

Is 80 samples.
[0061]
LP coefficient code conversion circuit 1100 receives the first LP coefficient code from code separation circuit 1010. Here, "3GPP \ AMR \ Speech \ Codec" (Reference 3) and ITU-T Recommendation G.264. In many standard schemes such as G.729, the LP coefficient is often represented by a line spectrum pair (Line Spectral Pair: LSP), and the LSP is often encoded and decoded. Therefore, the encoding and decoding of the LP coefficient are performed in the LSP domain. Suppose Regarding the conversion from the LP coefficient to the LSP, and the conversion from the LSP to the LP coefficient, reference is made to a well-known method, for example, the description in Section 5.2.3 and Section 5.2.4 of “Document 3”. . The LP coefficient code conversion circuit 1100 decodes the first LP coefficient code by the LSP decoding method in the scheme A to obtain a first LSP.
[0062]
Next, the LP coefficient code conversion circuit 1100 quantizes and encodes the first LSP according to the LSP quantization method and the encoding method in the method B, and generates a second LSP and a code corresponding thereto (the second LSP). LP coefficient sign). Then, the LP coefficient code conversion circuit 1100 outputs the second LP coefficient code to the code multiplexing circuit 1020 as a code that can be decoded by the LSP decoding method in the method B, and outputs the first LSP and the second LSP to the LSP. -Output to LPC conversion circuit 1110.
[0063]
FIG. 2 is a diagram showing a configuration of the LP coefficient code conversion circuit 1100. Referring to FIG. 2, the LP coefficient code conversion circuit 1100 includes an LSP decoding circuit 110, a first LSP codebook 111, an LSP coefficient coding circuit 130, and a second LSP codebook 131. Each component of LP coefficient code conversion circuit 1100 will be described with reference to FIG.
[0064]
The LSP decoding circuit 110 decodes the corresponding LSP from the LP coefficient code. The LSP decoding circuit 110 includes a first LSP codebook 111 in which a plurality of sets of LSPs are stored, and inputs the first LP coefficient code output from the code separation circuit 1010 via the input terminal 31. , The LSP corresponding to the first LP coefficient code is read from the first LSP codebook 111, the read LSP is output to the LSP encoding circuit 130 as the first LSP, and the LSP is output via the output terminal 33. -Output to LPC conversion circuit 1110. Here, decoding of the LSP from the LP coefficient code uses the LSP codebook of the system A according to the LSP decoding method of the system A.
[0065]
The LSP encoding circuit 130 receives the first LSP output from the LSP decoding circuit 110, and outputs a second LSP and a corresponding LP coefficient code from a second LSP codebook 131 storing a plurality of sets of LSPs. Are sequentially read, a second LSP that minimizes an error from the first LSP is selected, and a corresponding LP coefficient code is set as a second LP coefficient code via an output terminal 32 through a code multiplexing circuit 1020. And outputs the second LSP to the LSP-LPC conversion circuit 1110 via the output terminal 34. Here, the second LSP selection method, that is, the LSP quantization and encoding method uses the LSP codebook of the method B according to the LSP quantization method and the encoding method of the method B. Here, for the quantization and encoding of the LSP, for example, the description in Section 5.2.5 of “Document 3” is referred to.
[0066]
Thus, the description of the LP coefficient code conversion circuit 1100 shown in FIG. 2 is completed, and the description returns to FIG.
[0067]
The LSP-LPC conversion circuit 1110 receives the first LSP and the second LSP output from the LP coefficient code conversion circuit 1100, and converts the first LSP into a first LP coefficient a._{1, i}And convert the second LSP to a second LP coefficient a_{2, i}And the first LP coefficient a_{1, i}To the target signal calculation circuit 1700, the speech decoding circuit 1500, and the impulse response calculation circuit 1120, and the second LP coefficient a_{2, i}Is output to the target signal calculation circuit 1700 and the impulse response calculation circuit 1120. Here, regarding the conversion from the LSP to the LP coefficient, the description in Section 5.2.4 of “Document 3” is referred to.
[0068]
The ACB code conversion circuit 1200 obtains a second ACB code by reading the first ACB code input from the code separation circuit 1010 using the correspondence between the code in the method A and the code in the method B. Then, ACB code conversion circuit 1200 outputs the second ACB code to code multiplexing circuit 1020 as a code that can be decoded by the ACB decoding method in scheme B. Further, ACB code conversion circuit 1200 outputs an ACB delay corresponding to the second ACB code to target signal calculation circuit 1700 as a second ACB delay.
[0069]
Here, with reference to FIG. For example, ACB code in scheme A

Is 56, the corresponding ACB delay

Is 76. In scheme B, the ACB code

Is 53, the corresponding ACB delay

Is 76, in order to convert the ACB code from the method A to the method B so that the value of the ACB delay becomes the same (76 in this case), the ACB code 56 in the method A is converted into the ACB code in the method B. What is necessary is just to correspond to code | symbol 53. Thus, the description of the code replacement is completed, and the description returns to FIG.
[0070]
The audio decoding circuit 1500 receives the first ACB code, the first FCB code, and the first gain code output from the code separation circuit 1010, and receives the first LP coefficient from the LSP-LPC conversion circuit 1110. . Next, the audio decoding circuit 1500 uses the ACB signal decoding method, the FCB signal decoding method, and the gain decoding method in the method A to convert the first ACB code, the first FCB code, and the first gain code. From each of them, the ACB delay, the FCB signal and the gain are decoded, and each is set as a first ACB delay, a first FCB signal and a first gain. The audio decoding circuit 1500 generates an ACB signal using the first ACB delay, and sets this as the first ACB signal. Then, the audio decoding circuit 1500 generates audio from the first ACB signal, the first FCB signal, the first gain, and the first LP coefficient, and outputs the audio to the target signal calculation circuit 1700.
[0071]
FIG. 4 is a diagram showing a configuration of the audio decoding circuit 1500. Referring to FIG. 4, audio decoding circuit 1500 includes excitation signal information decoding circuit 1600 having ACB decoding circuit 1510, FCB decoding circuit 1520, gain decoding circuit 1530, excitation signal calculation circuit 1540, and excitation signal storage circuit. 1570 and a synthesis filter 1580. Referring to FIG. 4, each component of speech decoding circuit 1500 will be described.
[0072]
The excitation signal information decoding circuit 1600 decodes the information of the excitation signal from the code corresponding to the information of the excitation signal. The first ACB code, the first FCB code, and the first gain code output from the code separation circuit 1010 are input via input terminals 51, 52, and 53, respectively, and the first ACB code, the first FCB code, The ACB delay, the FCB signal, and the gain are decoded from the code and the first gain code, respectively, and are respectively defined as a first ACB delay, a first FCB signal, and a first gain. Here, the first gain is composed of an ACB gain and an FCB gain, and they are respectively referred to as a first ACB gain and a first FCB gain. Further, the excitation signal information decoding circuit 1600 receives the past excitation signal output from the excitation signal storage circuit 1570. The excitation signal information decoding circuit 1600 generates an ACB signal using the past excitation signal and the first ACB delay, and sets this as the first ACB signal. Then, the excitation signal information decoding circuit 1600 outputs the first ACB signal, the first FCB signal, the first ACB gain, and the first FCB gain to the excitation signal calculation circuit 1540.
[0073]
Next, the ACB decoding circuit 1510, FCB decoding circuit 1520, and gain decoding circuit 1530, which are components of the excitation signal information decoding circuit 1600, will be described in detail.
[0074]
The ACB decoding circuit 1510 inputs the first ACB code output from the code separation circuit 1010 via the input terminal 51, and inputs a past excitation signal output from the excitation signal storage circuit 1570. Next, similarly to ACB code conversion circuit 1200 described above, ACB decoding circuit 1510 uses the correspondence between the ACB code and ACB delay in method A shown in FIG. 3 to generate a first ACB code corresponding to the first ACB code. ACB delay

Get. In the excitation signal, from the start of the current subframe

Equivalent to subframe length, from a point in the sample past

A sample signal is cut out to generate a first ACB signal. here,

But

If smaller than

Cut out a vector for the sample, connect this vector repeatedly, and

Let it be a sample signal. Then, the first ACB signal is output to excitation signal calculation circuit 1540. Here, for details of the method of generating the first ACB signal, refer to the descriptions in Sections 6.1 and 5.6 of “Document 3”.
[0075]
The FCB decoding circuit 1520 receives the first FCB code output from the code separation circuit 1010 via the input terminal 52, and converts the first FCB signal corresponding to the first FCB code into an excitation signal calculation circuit 1540. Output to The FCB signal is represented by a multi-pulse signal defined by a pulse position and a pulse polarity, and the first FCB code is a code (pulse position code) corresponding to the pulse position and a code (pulse polarity code) corresponding to the pulse polarity. ). Here, for details of the method of generating the FCB signal represented by the multi-pulse signal, refer to the descriptions in Sections 6.1 and 5.7 of “Document 3”.
[0076]
The gain decoding circuit 1530 inputs the first gain code output from the code separation circuit 1010 via the input terminal 53. The gain decoding circuit 1530 incorporates a table in which a plurality of gains are stored, and reads a gain corresponding to the first gain code from the table. Then, the gain decoding circuit 1530 outputs the first ACB gain corresponding to the ACB gain and the first FCB gain corresponding to the FCB gain among the read gains to the excitation signal calculation circuit 1540. Here, when the first ACB gain and the first FCB gain are collectively encoded, the table stores a plurality of two-dimensional vectors including the first ACB gain and the first FCB gain. ing. When the first ACB gain and the first FCB gain are individually encoded, two tables are built in, and one table stores a plurality of first ACB gains and the other table stores the first ACB gain. A plurality of first FCB gains are stored in the table.
[0077]
Excitation signal calculation circuit 1540 receives the first ACB signal output from ACB decoding circuit 1510, receives the first FCB signal output from FCB decoding circuit 1520, and outputs the first FCB signal output from gain decoding circuit 1530. The first ACB gain and the first FCB gain are input. Excitation signal calculation circuit 1540 adds a signal obtained by multiplying the first ACB signal by the first ACB gain and a signal obtained by multiplying the first FCB signal by the first FCB gain to obtain a first signal. To obtain the excitation signal. Then, the excitation signal calculation circuit 1540 outputs the first excitation signal to the synthesis filter 1580 and the excitation signal storage circuit 1570.
[0078]
The excitation signal storage circuit 1570 receives the first excitation signal output from the excitation signal calculation circuit 1540, and stores and holds the first excitation signal. Then, the excitation signal storage circuit 1570 outputs the past first excitation signal that has been input and stored in the past to the ACB decoding circuit 1510.
[0079]
The synthesis filter 1580 inputs the first excitation signal output from the excitation signal calculation circuit 1540, and inputs the first LP coefficient output from the LSP-LPC conversion circuit 1110 via the input terminal 61. Then, the synthesis filter 1580 generates an audio signal by driving the linear prediction filter having the first LP coefficient with the first excitation signal. The audio signal is output to the target signal calculation circuit 1700 via the output terminal 63.
[0080]
This is the end of the description of the speech decoding circuit 1500 shown in FIG. 4, and returns to the description of FIG.
[0081]
The target signal calculation circuit 1700 receives the first LSP and the second LSP from the LSP-LPC conversion circuit 1110, and receives the second ACB delay corresponding to the second ACB code from the ACB code conversion circuit 1200. , A decoded speech from the speech decoding circuit 1500, an impulse response signal from the impulse response calculation circuit 1120, and a past second excitation signal stored and held in the second excitation signal storage circuit 1620. The target signal calculation circuit 1700 calculates a first target signal from the decoded speech, the first LP coefficient, and the second LP coefficient. Next, the target signal calculation circuit 1700 obtains a second ACB signal and an optimum ACB gain from the past second excitation signal, impulse response signal, first target signal, and second ACB delay. Then, the target signal calculation circuit 1700 outputs the first target signal and the optimum ACB gain to the gain code generation circuit 1400, and outputs the second ACB signal to the gain code generation circuit 1400 and the second excitation signal calculation circuit 1610. Output to
[0082]
FIG. 5 is a diagram showing a configuration of the target signal calculation circuit 1700. Referring to FIG. 5, the target signal calculation circuit 1700 includes a weighting signal calculation circuit 1710, an ACB signal generation circuit 1720, and an optimum ACB gain calculation circuit 1730. With reference to FIG. 5, each component of the target signal calculation circuit 1700 will be described.
[0083]
The weighting signal calculation circuit 1710 inputs the decoded speech s (n) output from the synthesis filter 1580 of the speech decoding circuit 1500 via the input terminal 57, and outputs the first LP coefficient output from the LSP-LPC conversion circuit 1110. a_{1, i}And the second LP coefficient a_{2, i}Are input via the input terminal 36 and the input terminal 35, respectively. The weighting signal calculation circuit 1710 first configures the audibility weighting filter W (z) using the first LP coefficient.
[0084]
Then, the weighting signal calculation circuit 1710 drives the audibility weighting filter with the decoded voice to generate the audibility weighted voice signal. Next, the weighting signal calculation circuit 1710 configures an auditory weighting synthesis filter W (z) / A2 (z) using the first LP coefficient and the second LP coefficient.
[0085]
The weighting signal calculation circuit 1710 subtracts the zero input response of the perceptual weighting synthesis filter from the perceptual weighting audio signal, and outputs the first target signal x (n) to the ACB signal generating circuit 1720 and the optimum ACB gain calculating circuit. The signal is output to the second target signal calculation circuit 1430 via the output terminal 78.
[0086]
The ACB signal generation circuit 1720 receives the first target signal output from the weighting signal calculation circuit 1710, and receives a second ACB delay T output from the ACB code conversion circuit 1200.^(B) _lagIs input via an input terminal 37, the impulse response signal h (n) output from the impulse response calculation circuit 1120 is input via an input terminal 74, and the past output from the second excitation signal storage circuit 1620 is input. The second excitation signal u (n) is input via the input terminal 75.
[0087]
The ACB signal generation circuit 1720 convolves the impulse response signal with a signal cut out from the past second excitation signal with a delay k, thereby filtering the past excitation signal with a delay k.

Is calculated.
[0088]
Here, the delay k is a second ACB delay. A signal cut out from the past second excitation signal with a delay k is defined as a second ACB signal v (n).
[0089]
Then, the ACB signal generation circuit 1720 outputs the second ACB signal to the second target signal calculation circuit 1430 and the second excitation signal calculation circuit 1610 via the output terminal 76, and outputs the filtered delay k The past excitation signal yk (n) is output to the optimum ACB gain calculation circuit 1730.
[0090]
The optimum ACB gain calculation circuit 1730 receives the first target signal x (n) output from the weighting signal calculation circuit 1710, and outputs the past excitation signal of the filtered delay k output from the ACB signal generation circuit 1720. Enter yk (n).
[0091]
Next, the optimum ACB gain calculation circuit 1730 calculates the optimum ACB gain gp from the first target signal x (n) and the past excitation signal yk (n) of the filtered delay k by the following equation. . The optimum ACB gain gp is a gain that minimizes the distance between the first target signal x (n) and the past excitation signal yk (n) with the filtered delay k.

[0092]
Then, the optimum ACB gain calculation circuit 1730 outputs the optimum ACB gain gp to the ACB gain encoding circuit 1410 via the output terminal 77.
[0093]
The details of the method of calculating the second ACB signal and the method of calculating the optimum ACB gain can be referred to the descriptions in Sections 6.1 and 5.6 of “Document 3”. This is the end of the description of the target signal calculation circuit 1700 in FIG. 5, and returns to the description of FIG.
[0094]
The impulse response calculation circuit 1120 receives the first LP coefficient and the second LP coefficient output from the LSP-LPC conversion circuit 1110, and uses the first LP coefficient and the second LP coefficient to perform an auditory weighting synthesis filter. Is composed.
[0095]
Then, impulse response calculation circuit 1120 outputs the impulse response signal of the perceptual weighting synthesis filter to target signal calculation circuit 1700 and gain code generation circuit 1400. Here, the transfer function of the perceptual weighting synthesis filter is represented by the following equation.

[0096]
However,

[0097]
Is the second LP coefficient

Is the transfer function of the linear prediction filter with.
[0098]

[0099]
Is the first LP coefficient

Is a transfer function of an auditory weighting filter having
[0100]
Here, P is a linear prediction order (for example, 10), and γ1 and γ2 are coefficients (for example, 0.94 and 0.6) for controlling weighting.
[0101]
The FCB code generation circuit 1800 receives the first FCB code output from the code separation circuit 1010, and converts the first FCB code into a code that can be decoded by the method B. The FCB code generation circuit 1800 outputs the converted FCB code as a second FCB code to the code multiplexing circuit 1020, and outputs a second FCB signal corresponding to the second FCB code to the gain code generation circuit 1400, 2 to the excitation signal calculation circuit 1610. Here, the FCB signal is composed of a plurality of pulses, and is represented by a multi-pulse signal defined by a pulse position (pulse position) and a polarity (pulse polarity). The FCB code includes a code corresponding to the pulse position (pulse position code) and a code corresponding to the pulse polarity (pulse polarity code). For the expression method of the FCB signal by the multi-pulse signal, refer to the description in Section 5.7 of “Document 3”.
[0102]
FIG. 6 is a diagram showing a configuration of the FCB code generation circuit 1800 in FIG. Referring to FIG. 6, the FCB code generation circuit 1800 includes an FCB code conversion circuit 1300 and an FCB signal generation circuit 1820. With reference to FIG. 6, each component of FCB code generation circuit 1800 will be described.
[0103]
The FCB code conversion circuit 1300 outputs the first FCB code i input from the code separation circuit 1010 via the input terminal 85.^(A) _PIs read using the correspondence between the code in the method A and the code in the method B to obtain the second FCB code i.^{( B )} _PGet. Then, the FCB code conversion circuit 1300 outputs this to the code multiplexing circuit 1020 via the output terminal 55 as a code decodable by the FCB decoding method in the system B, and outputs the pulse position corresponding to the second FCB code.

And pulse polarity

To the FCB signal generation circuit 1820.
[0104]
With reference to FIG. 7, the reading of the pulse position code will be described.
[0105]
For example, the pulse position code in scheme A

Is 6, the corresponding pulse position

Is 30. In method B, the pulse position code

Is 1, the corresponding pulse position

Is 30. In order to convert the pulse position code from scheme A to scheme B so that the pulse position value is the same (30 in this case), the pulse position code 6 in scheme A is converted to scheme B May be associated with the pulse position code 1 in.
[0106]
Regarding the pulse polarity code, the code may be read so that the polarity (positive or negative) corresponding to the code before reading and the polarity corresponding to the code after reading are equal.
[0107]
As described above, the description of the replacement of the pulse position code and the pulse polarity code is completed, and the description returns to FIG.
[0108]
The FCB signal generation circuit 1820 inputs the pulse position and the pulse polarity output from the FCB code conversion circuit 1300. The FCB signal generation circuit 1820 sets the FCB signal defined by the pulse position and the pulse polarity as the second FCB signal c (n), and outputs this to the optimum FCB gain calculation circuit 1440 and the second excitation signal calculation circuit 1610. Output via terminal 86.
[0109]
This is the end of the description of the FCB code generation circuit 1800 in FIG. 6, and returns to the description of FIG.
[0110]
The gain code generation circuit 1400 receives the first target signal, the second ACB signal, and the optimum ACB gain output from the target signal calculation circuit 1700, and receives the second FCB signal output from the FCB code generation circuit 1800. , The impulse response signal output from the impulse response calculation circuit 1120 is input, and the first LSP output from the LP coefficient code conversion circuit 1100 is input.
[0111]
The gain code generation circuit 1400 first calculates a second target signal from the first target signal, the second ACB signal, the optimum ACB gain, and the impulse response signal, and calculates the second target signal and the second FCB signal. And an impulse response signal, calculate a corrected FCB gain from the optimum FCB gain, and determine a voice determination value from the first LSP.
[0112]
Next, the gain code generation circuit 1400 calculates a first square error from the ACB gain sequentially read from the ACB gain codebook and the optimum ACB gain, and calculates a second square error from the ACB gain and the corrected ACB gain. I do.
[0113]
Then, the gain code generation circuit 1400 calculates the ACB gain and the corresponding ACB gain code that minimize the evaluation function calculated from the weight coefficient calculated from the speech determination value, the first square error, and the second square error. select.
[0114]
Further, the gain code generation circuit 1400 calculates a third square error from the FCB gain sequentially read from the FCB gain codebook and the optimal FCB gain, and calculates a fourth square error from the FCB gain and the corrected FCB gain. .
[0115]
Then, the gain code generation circuit 1400 calculates the FCB gain and the corresponding FCB gain code that minimize the evaluation function calculated from the weight coefficient calculated from the voice determination value, the third square error, and the fourth square error. select.
[0116]
Finally, the gain code generation circuit 1400 outputs the second gain code composed of the selected ACB gain code and FCB gain code to the code multiplexing circuit 1020 as a code that can be decoded by the gain decoding method in the scheme B. Output via.
[0117]
FIG. 8 is a diagram showing a configuration of the gain code generation circuit 1400. Referring to FIG. 8, a gain code generation circuit 1400 includes an ACB gain coding circuit 1410, an ACB gain codebook 1411, an FCB gain coding circuit 1420, an FCB gain codebook 1421, and a second target signal calculation circuit. 1430, an optimum FCB gain calculation circuit 1440, an optimum FCB gain correction circuit 1450, and a voice / non-voice discrimination circuit 1460. With reference to FIG. 8, each component of gain code generation circuit 1400 will be described in detail.
[0118]
The second target signal calculation circuit 1430 receives the second ACB signal v (n) output from the ACB signal generation circuit 1720 via the input terminal 92, and outputs the first ACB signal v (n) output from the weighting signal calculation circuit 1710. The target signal x (n) is input through the input terminal 93, the impulse response signal h (n) output from the impulse response calculation circuit 1120 is input through the input terminal 94, and the output from the ACB gain encoding circuit 1410 is output. Input second ACB gain.
[0119]
The second target signal calculation circuit 1430 generates a filtered second ACB signal by convolution of the second ACB signal and the impulse response signal.

And calculate the second ACB gain in y (n).

Is subtracted from the first target signal x (n) to obtain a second target signal x (n).₂(N) is obtained.

[0120]
Then, the second target signal calculation circuit 1430 calculates the second target signal x₂(N) is output to the optimal FCB gain calculation circuit 1440.
[0121]
The optimal FCB gain calculation circuit 1440 inputs the second FCB signal c (n) output from the FCB signal generation circuit 1820 through the input terminal 91, and outputs the impulse response signal h () output from the impulse response calculation circuit 1120. n) is input via the input terminal 94, and the second target signal x output from the second target signal calculation circuit 1430₂(N), the second FCB signal filtered by convolution of the second FCB signal and the impulse response signal

Is calculated from the second target signal x2 (n) and the filtered second FCB signal z (n) by the following equation. The optimal FCB gain gc is a gain that minimizes the distance between the second target signal x2 (n) and the filtered second FCB signal z (n).

[0122]
Then, the optimal FCB gain calculation circuit 1440 outputs the optimal FCB gain to the optimal FCB gain correction circuit 1450 and the FCB gain encoding circuit 1420.
[0123]
The voice / non-voice discriminating circuit 1460 inputs the first LSP output from the LSP decoding circuit 110 via the input terminal 98. An LSP fluctuation amount is calculated from the first LSP and its long-term average, and a voice determination value is determined from the LSP fluctuation amount.
[0124]
The procedure for obtaining the LSP fluctuation amount will be described below. In the n-th frame, the long-term average of the LSP

Is calculated by the following equation.

Here, Np is a linear prediction order, and β is, for example, 0.9.
[0125]
The variation dq (n) of the LSP in the n-th frame is defined by the following equation.

here,

Is

When

As an error with, for example,

Or

Can be defined, but the latter is used here. A section with a large variation dq (n) can correspond to a voice section, and a section with a small variation dq (n) can correspond to a non-voice section. By the threshold processing for the fluctuation amount dq (n), the sound determination value

To determine.
[0126]

(When Vs = 1 dq (n) is not less than CVS
Vs = 0 dq (n) is smaller than CVS)
[0127]
Here, Cvs is a certain constant (for example, 2.2), Vs = 1 corresponds to a voice section, and Vs = 0 corresponds to a non-voice section. The voice determination value is output to the optimum ACB gain correction circuit 1480, the ACB gain coding circuit 1410, the optimum FCB gain correction circuit 1450, and the FCB gain coding circuit 1420.
[0128]
The optimum ACB gain correction circuit 1480 inputs the optimum ACB gain output from the ACB signal generation circuit 1720 via the input terminal 97, and inputs the voice determination value output from the voice / non-voice recognition circuit 1460. When the voice determination value Vs is 0 (non-voice section), the optimum ACB gain correction circuit 1480 sets the long-term average of the optimum ACB gain as the corrected ACB gain. In a non-voice section, a long-term average of the optimum ACB gain is calculated by the following equation.

[0129]
here,

Is the optimal ACB gain in the n-th subframe,

Is the long-term average of the optimum ACB gain in the n-th subframe, and α is, for example, 0.9. Note that an average value, a median value, a mode value, and the like can be applied to the long-term average.
[0130]
On the other hand, the optimum ACB gain correction circuit 1480 sets the optimum ACB gain itself as the corrected ACB gain when the voice determination value Vs is 1 (voice section).
[0131]
Optimal ACB gain correction circuit 1480 outputs the corrected ACB gain to ACB gain encoding circuit 1410.
[0132]
The ACB gain encoding circuit 1410 inputs the optimum ACB gain gp output from the ACB signal generation circuit 1720 via the input terminal 97, inputs the corrected ACB gain output from the optimum ACB gain correction circuit 1480, and The voice judgment value output from the non-voice discriminating circuit 1460 is input.
[0133]
The ACB gain encoding circuit 1410 calculates a first squared error from the ACB gain sequentially read from the ACB gain codebook 1411 and the optimum ACB gain from the input terminal 97, and calculates a second square error from the ACB gain and the corrected ACB gain. The square error is calculated, and an evaluation function defined by the following equation is calculated from the weight coefficient calculated from the voice determination value, the first square error, and the second square error.

[0134]
here,

Is the optimal ACB gain,

Is the modified ACB gain,

Is an ACB gain sequentially read from the ACB gain codebook, and μ is a weight coefficient. For example, when the voice determination value Vs is 1 (voice section), the weight coefficient μ is 1.0, and when Vs is 0 (non-voice section), μ is 0.2.
[0135]
Then, the ACB gain encoding circuit 1410 selects an ACB gain that minimizes the evaluation function, outputs the selected ACB gain as a second ACB gain to the second target signal calculation circuit 1430, and The signal is output to the excitation signal calculation circuit 1610 via the output terminal 95, and the code corresponding to the second ACB gain is output to the gain code multiplexing circuit 1470 as the ACB gain code.
[0136]
The optimal FCB gain correction circuit 1450 receives the optimal FCB gain output from the optimal FCB gain calculation circuit 1440 and receives the audio determination value Vs output from the audio / non-voice discrimination circuit 1460.
[0137]
When the voice determination value Vs is 0 (non-voice section) in the optimum FCB gain correction circuit 1450, the long-term average of the optimum FCB gain is set as the corrected FCB gain. In a non-voice section, a long-term average of the optimal FCB gain is calculated by the following equation.

[0138]
here,

Is the optimal FCB gain in the n-th subframe,

Is the long-term average of the optimal FCB gain in the n-th subframe, and α is, for example, 0.9. Note that an average value, a median value, a mode value, and the like can be applied to the long-term average.
[0139]
On the other hand, in the optimum FCB gain correction circuit 1450, when the voice determination value Vs is 1 (voice section), the optimum FCB gain itself is set as a corrected FCB gain.
[0140]
Optimal FCB gain correction circuit 1450 outputs the corrected FCB gain to FCB gain encoding circuit 1420.
[0141]
The FCB gain encoding circuit 1420 receives the optimum FCB gain output from the optimum FCB gain calculation circuit 1440, the corrected FCB gain output from the optimum FCB gain correction circuit 1450, and receives the voice / non-voice discrimination circuit 1460. Enter the output voice judgment value. The FCB gain encoding circuit 1420 calculates a first square error from the FCB gain sequentially read from the FCB gain codebook 1421 and the optimal FCB gain, and calculates a second square error from the FCB gain and the corrected FCB gain. Then, an evaluation function defined by the following equation is calculated from the weight coefficient calculated from the voice determination value, the first square error, and the second square error.

[0142]
here,

Is the optimal FCB gain,

Is the modified FCB gain,

Is an FCB gain sequentially read from the FCB gain codebook, and μ is a weight coefficient. For example, when the voice determination value Vs is 1 (voice section), the weight coefficient μ is 1.0, and when the voice determination value Vs is 0 (non-voice section), μ is 0.2.
[0143]
Then, the FCB gain encoding circuit 1420 selects the FCB gain with the smallest evaluation function, and outputs the selected FCB gain as the second FCB gain to the second excitation signal calculation circuit 1610 via the output terminal 96. Then, a code corresponding to the second FCB gain is output to gain code multiplexing circuit 1470 as an FCB gain code.
[0144]
The gain code multiplexing circuit 1470 receives the ACB gain code output from the ACB gain coding circuit 1410, inputs the FCB gain code output from the FCB gain coding circuit 1420, and divides the ACB gain code and the FCB gain code. The second gain code obtained by multiplexing is output to the code multiplexing circuit 1020 via the output terminal 56 as a code that can be decoded by the gain decoding method in the scheme B.
[0145]
This is the end of the description of the gain code generation circuit 1400 in FIG. 8, and returns to the description of FIG.
[0146]
The second excitation signal calculation circuit 1610 receives the second ACB signal output from the target signal calculation circuit 1700, receives the second FCB signal output from the FCB code generation circuit 1800, and receives a gain code generation circuit. The second ACB gain and the second FCB gain output from 1400 are input. The second excitation signal calculation circuit 1610 adds a signal obtained by multiplying the second ACB signal by the second ACB gain and a signal obtained by multiplying the second FCB signal by the second FCB gain. To obtain a second excitation signal. Then, the second excitation signal is output to the second excitation signal storage circuit 1620.
[0147]
The second excitation signal storage circuit 1620 receives the second excitation signal output from the second excitation signal calculation circuit 1610, and stores and holds the second excitation signal. Then, the second excitation signal input and stored in the past is output to the target signal calculation circuit 1700. This concludes the description of the first embodiment of the present invention.
[0148]
Next, a second embodiment of the present invention will be described. FIG. 9 is a diagram showing the configuration of a second embodiment of the transcoder according to the present invention. In FIG. 9, the LP coefficient code conversion circuit 100 and the gain code conversion circuit 400 in FIG. 12 are replaced by an LP coefficient code conversion circuit 1100 and a gain code conversion circuit 2400, respectively. A connection is added to the conversion circuit 2400. In the following, description of elements that are the same as or equivalent to the elements shown in FIG. 12 will be omitted, and differences will be described.
[0149]
The LP coefficient code conversion circuit 1100 is the same as that in the first embodiment described with reference to FIG. However, the way of connection with other circuits is different, and the first LSP is output to the gain code conversion circuit 400.
[0150]
Gain code conversion circuit 2400 receives the first gain code output from code separation circuit 1010, and receives the first LSP output from LP coefficient code conversion circuit 1100.
[0151]
First, the gain code conversion circuit 2400 converts the first gain code from the first gain (the first ACB gain and the first FCB gain) obtained by decoding the first gain code by the gain decoding method in the scheme A, to the modified ACB gain. And a corrected FCB gain, and a voice determination value is determined from the first LSP.
[0152]
Next, the gain code conversion circuit 2400 calculates a first square error from the ACB gain sequentially read from the ACB gain codebook and the first ACB gain, and calculates a second square error from the ACB gain and the corrected ACB gain. Is calculated.
[0153]
The gain code conversion circuit 2400 calculates the ACB gain and the corresponding ACB gain that minimize the evaluation function calculated from the weight coefficient calculated from the speech determination value, the first square error, and the second square error. Select a sign.
[0154]
The gain code conversion circuit 2400 calculates a third square error from the FCB gain sequentially read from the FCB gain codebook and the first FCB gain, and calculates a fourth square error from the FCB gain and the corrected FCB gain. calculate. Then, the gain code conversion circuit 2400 calculates the FCB gain and the corresponding FCB gain code that minimize the evaluation function calculated from the weight coefficient calculated from the voice determination value, the third square error, and the fourth square error. select.
[0155]
Finally, the gain code conversion circuit 2400 outputs the second gain code including the selected ACB gain code and FCB gain code to the code multiplexing circuit 1020 as a code that can be decoded by the gain decoding method in the scheme B.
[0156]
FIG. 10 is a diagram showing a configuration of the gain code conversion circuit 2400 in FIG. Referring to FIG. 10, gain code conversion circuit 2400 includes a speech / non-speech identification circuit 1460, a gain code separation circuit 2490, an ACB gain decoding circuit 2470, an ACB gain codebook 2471, an ACB gain correction circuit 2440, ACB gain encoding circuit 2410, ACB gain codebook 1411, FCB gain decoding circuit 2480, FCB gain codebook 2481, FCB gain correction circuit 2450, FCB gain encoding circuit 2420, FCB gain codebook 1421 , A gain code multiplexing circuit 1470. With reference to FIG. 10, each component of the gain code conversion circuit 2400 of this embodiment will be described. In FIG. 10, a speech / non-speech discriminating circuit 1460 and a gain code multiplexing circuit 1470 are basically the same as the elements shown in FIG. 8, and a description thereof will be omitted below.
[0157]
The gain code separation circuit 2490 inputs the first gain code output from the code separation circuit 1010 via the input terminal 45, and codes corresponding to the ACB gain and the FCB gain from the first gain code, that is, the first gain code. The ACB gain code and the first FCB gain code are separated, the first ACB gain code is output to the ACB gain decoding circuit 2470, and the first FCB gain code is output to the FCB gain decoding circuit 2480.
[0158]
The ACB gain decoding circuit 2470 includes an ACB gain codebook 2471 in which a plurality of sets of ACB gains are stored. The ACB gain decoding circuit 2470 receives the first ACB gain code output from the gain code separation circuit 2490, and receives the first ACB gain code. The ACB gain corresponding to the code is read from the first ACB gain codebook 2471, and the read ACB gain is output to the ACB gain correction circuit 2440 as the first ACB gain and output to the ACB gain encoding circuit 2410. . Here, the decoding of the ACB gain from the ACB gain code uses the ACB gain codebook of the system A according to the ACB gain decoding method in the system A.
[0159]
The FCB gain decoding circuit 2480 includes an FCB gain codebook 2481 in which a plurality of sets of FCB gains are stored, receives the first FCB gain code output from the gain code separation circuit 2490, and The FCB gain corresponding to the code is read from the first FCB gain codebook 2481, and the read FCB gain is output to the FCB gain correction circuit 2450 as the first FCB gain and to the FCB gain encoding circuit 2420. . Here, the decoding of the FCB gain from the FCB gain code uses the FCB gain codebook of the system A according to the decoding method of the FCB gain in the system A.
[0160]
The ACB gain correction circuit 2440 receives the first ACB gain output from the ACB gain decoding circuit 2470, and receives the voice determination value output from the voice / non-voice discrimination circuit 1460. When the voice determination value Vs is 0 (non-voice section), the long-term average of the first ACB gain is set as the corrected ACB gain.
[0161]
The ACB gain correction circuit 2440 calculates the long-term average of the first ACB gain in the non-voice section by the following equation.

[0162]
here,

Is the first ACB gain in the nth subframe,

Is the long-term average of the first ACB gain in the n-th subframe, and α is, for example, 0.9. Note that an average value, a median value, a mode value, and the like can be applied to the long-term average.
[0163]
On the other hand, when the voice determination value Vs is 1 (voice section), the ACB gain correction circuit 2440 sets the first ACB gain itself as the corrected ACB gain.
[0164]
ACB gain correction circuit 2440 outputs the corrected ACB gain to ACB gain encoding circuit 2410.
[0165]
The FCB gain correction circuit 2450 receives the first FCB gain output from the FCB gain decoding circuit 2480, and receives the voice determination value output from the voice / non-voice discrimination circuit 1460.
[0166]
When the voice determination value Vs is 0 (non-voice section) in the FCB gain correction circuit 2450, the long-term average of the first FCB gain is set as the corrected FCB gain. In the non-voice section, the long term average of the first FCB gain is calculated by the following equation.

[0167]
here,

Is the first FCB gain in the nth subframe,

Is the long-term average of the first FCB gain in the n-th subframe, and α is, for example, 0.9. Note that an average value, a median value, a mode value, and the like can be applied to the long-term average.
[0168]
On the other hand, when the voice determination value Vs is 1 (voice section), the FCB gain correction circuit 2450 sets the first FCB gain itself as the corrected FCB gain.
[0169]
FCB gain correction circuit 2450 outputs the corrected FCB gain to FCB gain encoding circuit 2420.
[0170]
The ACB gain encoding circuit 2410 receives the first ACB gain output from the ACB gain decoding circuit 2470, receives the corrected ACB gain output from the ACB gain correction circuit 2440, and Enter the output voice judgment value.
[0171]
The ACB gain encoding circuit 2410 calculates a first square error from the ACB gain sequentially read from the ACB gain codebook 1411 and the first ACB gain, and calculates a second square error from the ACB gain and the corrected ACB gain. Then, an evaluation function defined by the following equation is calculated from the weight coefficient calculated from the voice determination value, the first square error, and the second square error.
[0172]

[0173]
here,

Is the first ACB gain,

Is the modified ACB gain,

Is an ACB gain sequentially read from the ACB gain codebook 1411, and μ is a weight coefficient. For example, when the voice determination value Vs is 1 (voice section), the weight coefficient μ is 1.0, and when Vs is 0 (non-voice section), μ is 0.2.
[0174]
Then, the ACB gain encoding circuit 2410 selects the ACB gain that minimizes the evaluation function, sets the selected ACB gain as the second ACB gain, and sets the code corresponding to the second ACB gain to the second ACB gain. Output to the gain code multiplexing circuit 1470 as a code.
[0175]
The FCB gain encoding circuit 2420 receives the first FCB gain output from the FCB gain decoding circuit 2480, inputs the corrected FCB gain output from the FCB gain correction circuit 2450, and receives the voice / non-voice discrimination circuit 1460. Enter the output voice judgment value.
[0176]
The FCB gain encoding circuit 2420 calculates a third square error from the FCB gain sequentially read from the FCB gain codebook 1421 and the first FCB gain, and calculates a fourth square error from the FCB gain and the corrected FCB gain. Then, an evaluation function defined by the following equation is calculated from the weight coefficient calculated from the voice determination value, the third square error, and the fourth square error.

[0177]
here,

Is the first FCB gain,

Is the modified FCB gain,

Is an FCB gain sequentially read from the FCB gain codebook 1421, and μ is a weight coefficient. For example, when the voice determination value Vs is 1 (voice section), the weight coefficient μ is 1.0, and when the voice determination value Vs is 0 (non-voice section), μ is 0.2.
[0178]
Then, the FCB gain encoding circuit 2420 selects the FCB gain that minimizes the evaluation function, sets the selected FCB gain as the second FCB gain, and sets the code corresponding to the second FCB gain to the second FCB gain. Output to the gain code multiplexing circuit 1470 as a code.
[0179]
The transcoder of each embodiment of the present invention described above may be realized by computer control such as a digital signal processor. FIG. 11 is a diagram schematically showing an apparatus configuration in a case where the code conversion processing of each of the above embodiments is implemented by a computer as a third embodiment of the present invention. In the computer 1 that executes the program read from the recording medium 6, the first code obtained by encoding the audio by the first encoding / decoding device can be decoded by the second encoding / decoding device. In performing the code conversion process of converting to a code, the recording medium 6 includes:
(A) a process of obtaining a first linear prediction coefficient from a first code string;
(B) a process of obtaining information on an excitation signal from the first code sequence;
(C) a process of obtaining an excitation signal from information of the excitation signal;
(D) a process of generating an audio signal by driving a filter having a first linear prediction coefficient with an excitation signal;
(E) a process of calculating a gain (optimum gain) that minimizes the distance between the second audio signal generated from the information obtained from the second code sequence and the first audio signal;
(F) a process for correcting the optimal gain;
(G) First square error is calculated from the corrected optimum gain (corrected optimum gain) and the gain read from the gain codebook in the second method, and the optimum gain and the gain read from the gain codebook are calculated. From the gain codebook by calculating the second squared error from the gain codebook, and selecting the gain that minimizes the evaluation function based on the first squared error and the second squared error from the gain codebook. ,
Is recorded. The program is read from the recording medium 6 to the memory 3 via the recording medium reading device 5 and the interface 4 and executed. The above-described program may be stored in a nonvolatile memory such as a flash memory such as a mask ROM, and the recording medium includes the nonvolatile memory, as well as a CD-ROM, an FD, a digital versatile disk (DVD), and a magnetic tape (MT). In addition to a medium such as a portable HDD, for example, when the program is transmitted from a server device to a computer by a communication medium, a wired or wireless communication medium carrying the program is also included.
[0180]
In the fourth embodiment of the present invention, in the computer 1 executing the program read from the recording medium 6, the first code obtained by encoding the audio by the first encoding / decoding device is converted into the second code. In performing the code conversion process of converting into the second code decodable by the encoding / decoding device, the recording medium 6 includes:
(A) decoding the gain information from the first code string;
(B) a process for correcting the decoded gain (decoding gain);
(C) First square error is calculated from the corrected decoding gain (corrected decoding gain) and the gain read from the gain codebook in the second method, and the decoding gain and the gain read from the gain codebook are calculated. From the gain codebook by calculating the second squared error from the gain codebook, and selecting the gain that minimizes the evaluation function based on the first squared error and the second squared error from the gain codebook. ,
Is recorded.
[0181]
Although the present invention has been described with reference to the above embodiment, the present invention is not limited to the configuration of the above embodiment, and a person skilled in the art within the scope of the claims set forth in the claims. Needless to say, various changes and modifications that could be made are included.
[0182]
【The invention's effect】
As described above, according to the present invention, there is an effect that deterioration of background noise sound quality in a non-speech section can be reduced.
[0183]
The reason is that, in the present invention, the information obtained from the first audio signal and the second code string obtained by driving the synthesis filter having the first linear prediction coefficient from the first code string with the excitation signal is used. An optimum gain is derived from the generated second audio signal, and the optimum gain is further corrected. Based on the corrected optimum gain, the optimum gain, and the gain read from the gain codebook in the second scheme, This is because the gain information in the code string of No. 2 is obtained, and in this case, the second gain is obtained using an evaluation function that reduces the time variation of the second gain in the non-voice section. The above effect is obtained by decoding gain information from the first code string, correcting the decoded gain, and based on the corrected decoding gain, and the decoding gain and the gain read from the gain codebook in the second scheme. , The gain information in the second code string is obtained, and the second gain is obtained by using an evaluation function that reduces the time variation of the second gain in the non-voice section. Can also be played.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a first embodiment of a code conversion apparatus according to the present invention.
FIG. 2 is a diagram showing a configuration of an LP coefficient code conversion circuit in the code conversion device according to the present invention.
FIG. 3 is a diagram illustrating a correspondence relationship between an ACB code and an ACB delay and a method of reading the ACB code.
FIG. 4 is a diagram showing a configuration of a speech decoding circuit of the transcoder according to the present invention.
FIG. 5 is a diagram showing a configuration of a target signal calculation circuit in the transcoder according to the present invention.
FIG. 6 is a diagram showing a configuration of an FCB code generation circuit in the code conversion device according to the present invention.
FIG. 7 is a diagram for explaining a correspondence relationship between a pulse position code and a pulse position and a method of reading an ACB code.
FIG. 8 is a diagram showing a configuration of a gain code generation circuit in the code conversion device according to the present invention.
FIG. 9 is a diagram showing a configuration of a second embodiment of the code conversion apparatus according to the present invention.
FIG. 10 is a diagram showing a configuration of a gain code generation circuit in the code conversion device according to the present invention.
FIG. 11 is a diagram showing a configuration of a third to a fourth embodiment of the transcoder according to the present invention.
FIG. 12 is a diagram illustrating a configuration of a conventional transcoder.
[Explanation of symbols]
1 Computer
2 CPU
3 memory
4 Recording medium reading device interface
5 Recording medium reading device
6 Recording media
10, 31, 35, 36, 37, 51, 52, 53, 57, 61, 74, 75, 81, 82, 83, 84, 85, 91, 92, 93, 94 input terminals
20, 32, 33, 34, 55, 56, 62, 63, 76, 77, 78, 86, 95, 96 output terminals
100,1100 @ LP coefficient sign conversion circuit
110 LP coefficient decoding circuit
130 LP coefficient coding circuit
111 @ First LSP Codebook
131 @ 2nd LSP Codebook
200,1200 ACB code conversion circuit
300, 1300 {FCB} code conversion circuit
400, 2400 gain code conversion circuit
1010 code separation circuit
1020 code multiplexing circuit
1110 LSP-LPC conversion circuit
1120 ° impulse response calculation circuit
1400 gain code generation circuit
1410, 2410 ACB gain encoding circuit
1411,247 @ ACB Gain Codebook
1420, 2420 FCB gain coding circuit
1421,481 @ FCB Gain Codebook
1430 second target signal calculation circuit
1440 Optimal FCB gain calculation circuit
1450 ° optimal FCB gain correction circuit
1460 voice / non-voice discrimination circuit
1470 gain code multiplexing circuit
1480 Optimal ACB gain correction circuit
1500 audio decoding circuit
1510 @ ACB decoding circuit
1520 FCB decoding circuit
1530 gain decoding circuit
1540 excitation signal calculation circuit
1570 ° excitation signal storage circuit
1580 synthesis filter
1600 ° excitation signal information decoding circuit
1610 second excitation signal calculation circuit
1620 second excitation signal storage circuit
1700 target signal calculation circuit
1710 weighted signal calculation circuit
1720 ACB signal generation circuit
1800 FCB code generation circuit
1820 FCB signal generation circuit
2480 FCB gain decoding circuit
2450 FCB gain correction circuit
2490 gain code separation circuit

Claims (30)

In a code conversion method for converting a first code string conforming to the first method into a second code string conforming to the second method,
By obtaining information of a first linear prediction coefficient and an excitation signal from the first code string and driving a filter having the first linear prediction coefficient with an excitation signal obtained from the information of the excitation signal, the first Generating an audio signal of
Deriving an optimal gain based on a second audio signal generated by information obtained from a second code sequence and the first audio signal;
Modifying the optimal gain;
Obtaining gain information in a second code sequence based on the corrected optimal gain (referred to as “modified optimal gain”), the optimal gain, and a gain read from a gain codebook in the second method;
A code conversion method comprising: In a code conversion method for converting a first code string conforming to the first method into a second code string conforming to the second method,
Decoding gain information from the first code sequence;
Modifying the decoded gain (referred to as "decoding gain");
Obtaining gain information in a second code sequence based on a corrected decoding gain (referred to as “corrected decoding gain”), the decoding gain, and a gain read from a gain codebook in a second method;
A code conversion method comprising: Calculating a first squared error from the corrected optimal gain and a gain read from the gain codebook;
Calculating a second squared error from the optimal gain and a gain read from the gain codebook;
Obtaining gain information in a second code sequence by selecting a gain that minimizes an evaluation function based on the first square error and the second square error from the gain codebook;
The code conversion method according to claim 1, comprising: Calculating a first squared error from the modified decoding gain and a gain read from the gain codebook;
Calculating a second squared error from the decoding gain and a gain read from the gain codebook;
Obtaining gain information in a second code string by selecting a gain that minimizes an evaluation function based on the first square error and the second square error from the gain codebook;
The code conversion method according to claim 2, comprising: The code conversion method according to claim 1, wherein the corrected optimal gain is based on a long-term average of the optimal gain. The code conversion method according to claim 2, wherein the modified decoding gain is based on a long-term average of the decoding gain. The gain that minimizes a distance between a second audio signal generated based on information obtained from the second code sequence and the first audio signal is obtained as the optimum gain. 2. The code conversion method according to 1. The code conversion method according to any one of claims 3 to 7, wherein the evaluation function includes the first square error, the second square error, and a weight coefficient. In a code conversion device that converts a first code string compliant with the first method into a second code string compliant with the second method,
By obtaining information of a first linear prediction coefficient and an excitation signal from the first code string and driving a filter having the first linear prediction coefficient with an excitation signal obtained from the information of the excitation signal, the first An audio decoding circuit that generates an audio signal of
An optimum gain calculation circuit that calculates an optimum gain based on a second audio signal generated based on information obtained from a second code string and the first audio signal;
An optimum gain correction circuit for correcting the optimum gain,
A gain coding circuit for obtaining gain information in a second code sequence based on a corrected optimum gain (referred to as “corrected optimum gain”), the optimum gain, and a gain read from a gain codebook in the second method. When,
A transcoding device comprising: In a code conversion device that converts a first code string compliant with the first method into a second code string compliant with the second method,
A gain decoding circuit for decoding gain information from the first code string;
A decoding gain correction circuit for correcting the decoded gain (referred to as “decoding gain”);
Gain coding circuit for obtaining gain information in a second code sequence based on a corrected decoding gain (referred to as “corrected decoding gain”), the decoding gain, and a gain read from a gain codebook in the second method. ,
A transcoding device comprising: The gain encoding circuit,
Calculating a first squared error from the corrected optimal gain and the gain read from the gain codebook; calculating a second squared error from the optimal gain and the gain read from the gain codebook; Means for obtaining gain information in the second code sequence by selecting a gain that minimizes an evaluation function based on the first squared error and the second squared error from the gain codebook,
The transcoder according to claim 9, wherein: The gain encoding circuit,
Calculating a first squared error from the corrected decoding gain and the gain read from the gain codebook; calculating a second squared error from the decoded gain and the gain read from the gain codebook; Means for obtaining gain information in a second code sequence by selecting a gain that minimizes an evaluation function based on the first square error and the second square error from the gain codebook. The transcoder according to claim 10, wherein: The transcoder according to claim 9, wherein the corrected optimal gain is based on a long-term average of the optimal gain. 13. The transcoder according to claim 10, wherein the modified decoding gain is based on a long-term average of the decoding gain. The optimal gain calculation circuit outputs a gain that minimizes a distance between a second audio signal generated based on information obtained from the second code string and the first audio signal as the optimal gain, The transcoder according to claim 9, wherein: The code conversion device according to claim 10, wherein the evaluation function includes the first square error, the second square error, and a weight coefficient. A computer that constitutes a code conversion device that converts a first code string compliant with the first method into a second code string compliant with the second method,
(A) Obtaining information of a first linear prediction coefficient and an excitation signal from the first code string, and driving a filter having the first linear prediction coefficient with an excitation signal obtained from the information of the excitation signal. Generating a first audio signal by
(B) a process of calculating an optimal gain based on a second audio signal generated based on information obtained from a second code sequence and the first audio signal;
(C) processing for correcting the optimum gain;
(D) a process of obtaining gain information in a second code string based on a corrected optimum gain (referred to as “corrected optimum gain”), the optimum gain, and a gain read from a gain codebook in the second method. ,
The program to execute. A computer that constitutes a code conversion device that converts a first code string compliant with the first method into a second code string compliant with the second method,
(A) decoding gain information from the first code string;
(B) processing for correcting the decoded gain (referred to as “decoding gain”);
(C) a process of obtaining gain information in a second code string based on a corrected decoding gain (referred to as “corrected decoding gain”), the decoding gain, and a gain read from a gain codebook in the second method. ,
The program to execute. The program according to claim 17,
Calculating a first squared error from the corrected optimal gain and the gain read from the gain codebook; calculating a second squared error from the optimal gain and the gain read from the gain codebook; Causing the computer to execute a process of obtaining gain information in a second code string by selecting a gain that minimizes an evaluation function based on the first square error and the second square error from the gain codebook. Program. The program according to claim 18,
Calculating a first squared error from the corrected decoding gain and the gain read from the gain codebook; calculating a second squared error from the decoded gain and the gain read from the gain codebook; Causing the computer to execute a process of obtaining gain information in a second code string by selecting a gain that minimizes an evaluation function based on the first square error and the second square error from the gain codebook. Program. The program according to claim 17 or 19,
The program wherein the corrected optimal gain is based on a long-term average of the optimal gain. The program according to claim 18, wherein
The program wherein the modified decoding gain is based on a long-term average of the decoding gain. The program according to any one of claims 18 to 22,
Causing the computer to execute, as the optimum gain, a process of obtaining a gain that minimizes a distance between a second audio signal generated based on information obtained from the second code string and the first audio signal. Program. The program according to any one of claims 17 to 22,
A program, wherein the evaluation function comprises the first square error, the second square error, and a weight coefficient. A recording medium recording the program according to any one of claims 17 to 23. Code string data obtained by multiplexing a code obtained by encoding an audio signal by the first method is input to a code separation circuit, and based on the code separated by the code separation circuit, a different code string from the first method is used. A code conversion device that converts the converted code into a code conforming to the second system, supplies the converted code to a code multiplexing circuit, and outputs code string data obtained by multiplexing the converted code from the code multiplexing circuit. ,
A circuit that generates first and second linear prediction coefficients obtained by decoding in a first method and a second method based on the linear prediction coefficient code separated by the code separation circuit;
The adaptive codebook (ACB) code of the first system input from the code separation circuit is read by using the correspondence between the code of the first system and the code of the second system, thereby obtaining the ACB of the second system. An adaptive codebook code conversion circuit (“ACB code”) including means for obtaining a code, outputting the code to the code multiplexing circuit, and outputting an ACB delay corresponding to the second ACB code to the target signal calculation circuit as a second ACB delay. Conversion circuit))
Excitation signal information including an ACB code, a fixed codebook (FCB) code, and a gain code in the first scheme separated by the code separation circuit is received as an input, and each is decoded, and the linear signal separated by the code separation circuit is decoded. By driving a synthesis filter having a first linear prediction coefficient obtained by decoding in a first method based on the prediction coefficient code with an excitation signal obtained from the excitation signal information, a decoded speech signal is synthesized and output. An audio decoding circuit;
A FCB code of a first system output from the code separation circuit is input, the FCB code is converted into a code that can be decoded by a second system, and the converted FCB code is used as a second FCB code. A fixed codebook code generation circuit (referred to as “FCB code generation circuit”) that outputs to the code multiplexing circuit and outputs a second FCB signal corresponding to the second FCB code;
An impulse response calculation circuit that outputs an impulse response signal of an auditory weighting synthesis filter composed of the first linear prediction coefficient and the second linear prediction coefficient;
The target signal calculation circuit;
A gain code generation circuit;
With
The target signal calculation circuit,
A decoded speech output from the synthesis filter of the speech decoding circuit is input, and an auditory weighting filter configured using the first linear prediction coefficient is driven by the decoded speech to generate an auditory weighted speech signal, A weighting signal calculation circuit for generating a first target signal obtained by subtracting the quiescent response of the audibility weighting synthesis filter configured using the first and second linear prediction coefficients from the audibility weighted speech signal; ,
The first target signal output from the weighting signal calculation circuit, the second ACB delay output from the ACB code conversion circuit, the impulse response signal output from the impulse response calculation circuit, And a past second excitation signal output from a second excitation signal storage circuit that stores and holds the second excitation signal of the above, and a delay k (where k is An ACB signal generation circuit that calculates a past excitation signal of a filtered delay k by convolution of the signal cut out by the second ACB delay) and the impulse response signal, and outputs the same as a second ACB signal; ,
Inputting the first target signal output from the weighting signal calculation circuit and the past excitation signal of the filtered delay k output from the ACB signal generation circuit, the first target signal An optimal ACB gain calculating circuit for deriving and outputting an optimal ACB gain from the past excitation signal of the filtered delay k,
With
The gain code generation circuit includes:
The first target signal, the second ACB signal, the optimum ACB gain, the second FCB signal output from the FCB code generation circuit, and the first target signal output from the target signal calculation circuit; Inputting the impulse response signal output from the impulse response calculation circuit and the first linear prediction coefficient,
Calculating a second target signal from the first target signal, the second ACB signal, the optimum ACB gain, and the impulse response signal, and calculating the second target signal, the second FCB signal, Means for calculating an optimal FCB gain from the impulse response signal;
Means for obtaining a corrected ACB gain from the optimum ACB gain;
Means for inputting the calculated optimal FCB gain and calculating a corrected FCB gain from the optimal FCB gain;
Means for determining a speech determination value from the first linear prediction coefficient;
Means for calculating a first squared error from an ACB gain sequentially read from an ACB gain codebook and the optimum ACB gain, and calculating a second squared error from the ACB gain and the corrected ACB gain;
An ACB gain and a corresponding ACB gain code that minimize a first evaluation function calculated from the weight coefficient calculated from the speech determination value, the first square error, and the second square error are selected. Means,
Means for calculating a third squared error from the FCB gain sequentially read from the FCB gain codebook and the optimum FCB gain, and calculating a fourth squared error from the FCB gain and the corrected FCB gain;
Means for selecting an FCB gain and a corresponding FCB gain code that minimize the second evaluation function calculated from the weight coefficient, the third square error, and the fourth square error calculated from the voice determination value;
Means for outputting a second gain code composed of the selected ACB gain code and FCB gain code as a code decodable by the gain decoding method in the second method, to the code multiplexing circuit;
A transcoder comprising: A second ACB signal output from the target signal calculation circuit; a second FCB signal output from the FCB code generation circuit; a second ACB gain output from the gain code generation circuit; An FCB gain is input, a signal obtained by multiplying the second ACB signal by a second ACB gain, and a signal obtained by multiplying the second FCB signal by a second FCB gain are added. A second excitation signal calculation circuit that obtains a second excitation signal through the second excitation signal and outputs the second excitation signal to the second excitation signal storage circuit.
The second excitation signal storage circuit inputs the second excitation signal output from the second excitation signal calculation circuit, stores and holds the second excitation signal, and stores the second excitation signal that has been input and stored in the past. 27. The transcoder according to claim 26, wherein an excitation signal is output to the target signal calculation circuit. The gain code generation circuit includes:
The second ACB signal output from the ACB signal generation circuit, the first target signal output from the weighting signal calculation circuit, the impulse response signal output from the impulse response calculation circuit, Receiving the second ACB gain output from the ACB gain encoding circuit, calculating a filtered second ACB signal by convolution of the second ACB signal and the impulse response signal, A signal obtained by multiplying the filtered second ACB signal by the second ACB gain is subtracted from the first target signal to derive a second target signal and output the second target signal. A second target signal calculation circuit,
The second FCB signal output from the FCB signal generation circuit, the impulse response signal output from the impulse response calculation circuit, and the second target signal output from the second target signal calculation circuit And calculates the filtered second FCB signal by convolution of the second FCB signal and the impulse response signal, and calculates the distance between the second target signal and the second FCB signal. An optimum FCB gain calculation circuit for calculating an optimum FCB gain to be minimized,
A speech / non-speech discrimination circuit that calculates a variation amount of the linear prediction coefficient from the first linear prediction coefficient and its long-term average to determine a speech determination value;
The optimum ACB gain output from the ACB signal generation circuit and the voice determination value output from the voice / non-voice discrimination circuit are input, and when the voice determination value is in a non-voice section, the optimum ACB gain is An optimal ACB gain correction circuit that calculates a long-term average of the optimum ACB gain in a non-voice section as a corrected ACB gain in a non-voice section, and outputs the optimum ACB gain itself as a corrected ACB gain in a voice section;
The optimum ACB gain output from the ACB signal generation circuit, the corrected ACB gain output from the optimum ACB gain correction circuit, and the voice determination value output from the voice / non-voice identification circuit are input. Calculating a first squared error from the ACB gain sequentially read from the ACB gain codebook and the optimum ACB gain, calculating a second squared error from the ACB gain and the corrected ACB gain, An evaluation function is obtained from a weight coefficient calculated from a determination value, the first square error, and the second square error, and an ACB gain that minimizes the evaluation function is selected, and the selected ACB gain is selected. As a second ACB gain to the second target signal calculation circuit, and to the second excitation signal calculation circuit to output the second ACB gain. And ACB code gain encoding circuit for outputting a code corresponding to the B gain to the gain code multiplexing circuit as ACB gain code,
The optimum FCB gain output from the optimum FCB gain calculation circuit and the voice determination value output from the voice / non-voice discrimination circuit are input. When the voice determination value is in a non-voice section, the optimum FCB gain is output. An optimal FCB gain correction circuit that outputs a corrected FCB gain to an FCB gain encoding circuit when the long-term average of the gain is a corrected FCB gain, and when the voice determination value is a voice section, the optimum FCB gain itself is a corrected FCB gain; ,
The optimum FCB gain output from the optimum FCB gain calculation circuit, the corrected FCB gain output from the optimum FCB gain correction circuit, and the voice determination value output from the voice / non-voice identification circuit are input. Calculating a third squared error from the FCB gain sequentially read from the FCB gain codebook and the optimal FCB gain, calculating a fourth squared error from the FCB gain and the corrected FCB gain, An evaluation function is calculated from the weight coefficient calculated from the determination value, the third square error, and the fourth square error, an FCB gain that minimizes the evaluation function is selected, and the selected FCB gain is calculated. The signal is output to the second excitation signal calculation circuit as a second FCB gain, and a code corresponding to the second FCB gain is used as an FCB gain code. And FCB gain encoding circuit to output to the in code multiplexing circuit,
A second ACB gain code obtained by inputting an ACB gain code output from the ACB gain coding circuit and an FCB gain code output from the FCB gain coding circuit, and multiplexing the ACB gain code and the FCB gain code. A gain code multiplexing circuit that outputs a gain code to the code multiplexing circuit as a code that can be decoded by the gain decoding method in the second method;
The transcoder according to claim 26, comprising: Code string data obtained by multiplexing a code obtained by encoding an audio signal by the first method is input to a code separation circuit, and based on the code separated by the code separation circuit, a different code string from the first method is used. A code conversion device that converts the converted code into a code conforming to the second system, supplies the converted code to a code multiplexing circuit, and outputs code string data obtained by multiplexing the converted code from the code multiplexing circuit. ,
A circuit that generates first and second linear prediction coefficients obtained by decoding in a first method and a second method based on the linear prediction coefficient code separated by the code separation circuit;
A first ACB code output from the code separation circuit is input, the first ACB code is converted into a code decodable by a second method, and the converted ACB code is used as a second ACB code. An ACB code conversion circuit for outputting to the code multiplexing circuit;
A first FCB code output from the code separation circuit is input, the first FCB code is converted into a code that can be decoded by a second method, and the converted FCB code is used as a second FCB code. An FCB code conversion circuit for outputting to the code multiplexing circuit;
A first gain code output from the code separation circuit is input, the first gain code is converted into a code that can be decoded by a second method, and the converted gain code is used as a second gain code. A gain code conversion circuit that outputs to the code multiplexing circuit;
With
The gain code conversion circuit,
A first gain code output from the code separation circuit and the first linear prediction coefficient are input, and a first gain code obtained by decoding the first gain code by a gain decoding method in a first scheme is obtained. Means for calculating a modified ACB gain and a modified FCB gain from the one adaptive codebook (ACB) gain and the first fixed codebook (FCB) gain;
Means for determining a speech determination value from the first linear prediction coefficient;
Calculating a first squared error from the ACB gain sequentially read from the ACB gain codebook and the first ACB gain; calculating a second squared error from the ACB gain and the corrected ACB gain; Means for selecting an ACB gain and a corresponding ACB gain code that minimize a first evaluation function calculated from a weight coefficient calculated from a determination value, the first square error, and the second square error When,
Calculating a third squared error from the FCB gain sequentially read from the FCB gain codebook and the first FCB gain; calculating a fourth squared error from the FCB gain and the corrected FCB gain; Means for selecting an FCB gain and a corresponding FCB gain code that minimize a second evaluation function calculated from the weighting factor calculated from the third squared error and the fourth squared error, and
Means for outputting to the code multiplexing circuit a second gain code composed of the selected ACB gain code and the FCB gain code as a code decodable by a gain decoding method in a second method;
A transcoder comprising: The gain code conversion circuit,
A speech / non-speech discrimination circuit that calculates a variation amount of the linear prediction coefficient from the first linear prediction coefficient and its long-term average to determine a speech determination value;
A first gain code output from the code separation circuit is input, and a first ACB gain code and a first FCB gain code corresponding to an ACB gain and an FCB gain are separated from the first gain code, and the first gain code is separated from the first gain code. A gain code separating circuit that outputs the ACB gain code to the ACB gain decoding circuit and outputs the first FCB gain code to the FCB gain decoding circuit;
An ACB gain codebook in which a plurality of sets of ACB gains are stored, a first ACB gain code output from the gain code separation circuit is input, and an ACB gain corresponding to the first ACB gain code is input. The ACB gain is read from the first ACB gain codebook, and the read ACB gain is output to the ACB gain correction circuit as the first ACB gain, and is also output to the ACB gain encoding circuit to decode the ACB gain from the ACB gain code. An ACB gain decoding circuit that uses the ACB gain codebook of the first scheme according to the ACB gain decoding method of the first scheme;
An FCB gain codebook storing a plurality of sets of FCB gains is provided, a first FCB gain code output from the gain code separation circuit is input, and an FCB gain corresponding to the first FCB gain code is input. Reading from the first FCB gain codebook, outputting the read FCB gain to the FCB gain correction circuit as the first FCB gain, and outputting the FCB gain to the FCB gain encoding circuit, and decoding the FCB gain from the FCB gain code Is an FCB gain decoding circuit that uses the FCB gain codebook of the first method according to the FCB gain decoding method of the first method;
The first ACB gain output from the ACB gain decoding circuit and the voice determination value output from the voice / non-voice identification circuit are input, and when the voice determination value is in a non-voice section, An ACB gain correction circuit that outputs the corrected ACB gain to an ACB gain encoding circuit while using the first ACB gain itself as a corrected ACB gain during a voice period; ,
The first FCB gain output from the FCB gain decoding circuit and the voice determination value output from the voice / non-voice identification circuit are input, and when the voice determination value is in a non-voice section, A long-term average of the FCB gains of 1 is a modified FCB gain, and when the voice judgment value is in a voice section, the first FCB gain itself is a modified FCB gain, and the modified FCB gain is output to an FCB gain encoding circuit. FCB gain correction circuit,
The first ACB gain output from the ACB gain decoding circuit, the corrected ACB gain output from the ACB gain correction circuit, and a voice determination value output from the voice / non-voice identification circuit are input. Calculating a first squared error from the ACB gain sequentially read from the ACB gain codebook and the first ACB gain, calculating a second squared error from the ACB gain and the corrected ACB gain, A first evaluation function is calculated from the weight coefficient calculated from the determination value, the first square error, and the second square error, and an ACB gain that minimizes the first evaluation function is selected. The selected ACB gain as a second ACB gain, and a code corresponding to the second ACB gain as a second ACB gain code to a gain code multiplexing circuit. And ACB gain encoding circuit for force,
The first FCB gain output from the FCB gain decoding circuit, the corrected FCB gain output from the FCB gain correction circuit, and the voice determination value output from the voice / non-voice identification circuit are input. Calculating a third squared error from the FCB gain sequentially read from the FCB gain codebook and the first FCB gain; calculating a fourth squared error from the FCB gain and the corrected FCB gain; A second evaluation function is calculated from the weight coefficient calculated from the voice determination value, the third squared error, and the fourth squared error, and an FCB gain that minimizes the second evaluation function is selected. Then, the selected FCB gain is set as a second FCB gain, and a code corresponding to the second FCB gain is set as a second FCB gain code to the gain code multiplexing circuit. And FCB gain encoding circuit to force,
A second ACB gain code obtained by inputting an ACB gain code output from the ACB gain coding circuit and an FCB gain code output from the FCB gain coding circuit, and multiplexing the ACB gain code and the FCB gain code. A gain code multiplexing circuit that outputs a gain code to the code multiplexing circuit as a code that can be decoded by the gain decoding method in the second method;
30. The transcoder according to claim 29, comprising:

Images